econstorMake Your Publications Visible.
A Service of
zbwLeibniz-InformationszentrumWirtschaftLeibniz Information Centrefor Economics
Schultz, T. Paul
Working Paper
An Economic Interpretation of the Decline in Fertilityin a Rapidly Developing Country: Consequences ofDevelopment and Family Planning
Center Discussion Paper, No. 258
Provided in Cooperation with:Yale University, Economic Growth Center (EGC)
Suggested Citation: Schultz, T. Paul (1977) : An Economic Interpretation of the Decline inFertility in a Rapidly Developing Country: Consequences of Development and Family Planning,Center Discussion Paper, No. 258, Yale University, Economic Growth Center, New Haven, CT
This Version is available at:http://hdl.handle.net/10419/160185
Standard-Nutzungsbedingungen:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichenZwecken und zum Privatgebrauch gespeichert und kopiert werden.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielleZwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglichmachen, vertreiben oder anderweitig nutzen.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten,gelten abweichend von diesen Nutzungsbedingungen die in der dortgenannten Lizenz gewährten Nutzungsrechte.
Terms of use:
Documents in EconStor may be saved and copied for yourpersonal and scholarly purposes.
You are not to copy documents for public or commercialpurposes, to exhibit the documents publicly, to make thempublicly available on the internet, or to distribute or otherwiseuse the documents in public.
If the documents have been made available under an OpenContent Licence (especially Creative Commons Licences), youmay exercise further usage rights as specified in the indicatedlicence.
www.econstor.eu
ECONOMIC GROWTH CENTER
YALE UNIVERSITY
Box 1987, Yale Station New Haven, Connecticut
CENTER DISCUSSION PAPER NO. 258
AN ECONOMIC INTERPRETATION OF THE DECLINE IN FERTILITY
IN A RAPIDLY DEVELOPING COUNTRY:
CONSEQUENCES OF DEVELOPMENT AND FAMILY PLANNING
T. Paul Schultz
March 1977
Note: Center Discussion·Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to Discussion Papers should be _cleared with the ~uthor to protect the tentative character of these papers.
... .: . _;__ ,:~ ~ . .,. - .:~ ~·-
Acknowledgments
This research was supported in part by AID contract otr-1432 and
facilitated by the Rockefeller Foundation grant RF 70051 to Yale's Economic
Demography Program. In the acquisition of data for Taiwan, I am grateful
for the help of A.I. Hermalin and T.H. Sun of the University of Michigan
and the Taiwan Provincial Institute of Family Planning, respectively. The
research assistance of N. Kwan, J. Oder and R. Beach is much appreciated,
I have benefited from the comments of B. Boulier, G. Fields, R. Freedman,
J. McCabe, T.W. Schultz, B. Tabbarah, and F. Welch.
AN ECONOMIC INTERPRETATION OF THE DECLlliE lli FERTILITY IN' A RAPIDLY DEVELOPlliG
COUNTRY: CONSEQUENCES OF DEVELOPMENT AND FAMILY PLANNING
Abstract
Two aspects of reproductive behavior are investigated for 21 regions
of Taiwan. A conventional economic model of fertility is fit to (1) the
logarithm of cohort fertility in 1966, (2) the logarithm of marriage
duration, and (3) the logarithm of the birth rate per year since first
marriage. The close association noted between regional levels of child
mortality and cohort fertility in Taiwan is linked with earlier marriage
in regions of high child mortality. Educational attairunent of men and women,
which are thought to affect parent reproductive goals, account for variation
in birth rates within marriage. A simple stock adjustment is then adapted
to explain both the stocks of births and subsequent flows of births to
women of various birth cohorts. The stocks in 1966 and 1971 and the pooled
flows from 1967 to 1974 imply similar estimates of the underlying reproduc-
tive. process: child mortality increases births (a replacement response),
female secondary schooling decreases births (predominantly a price of
mother's time effect), male secondary schooling increases births (predom-
inantly an income effect), and local family planning field worker activity
decreases births after a woman reaches age 30 (reduced informational and
monetary costs of birth control).
,:. .. ... - .:. ~ ..
I. INTRODUCTION
Parents make sacrifices to rear children. And though some rewards
of parenthood are virtually immediate, other benefits from having children
aannot be realized for years or even decades. In understanding the process
by which reproductive goals change, the demand for children should be inter-
preted, therefore, as in part a demand for a durable input that enters into
many lifetime production and consumption possibilities. 1 ·Given the number
of children parents want, the spacing of them undoubtedl~ confers on parents
relative advantages that might be explained in terms of either their life
cycle production and investment environment, or their anticipated psychologi-
cal and economic "returns to scale" in rearing of children at different time
intervals. 2 But as yet few theoretical insights have emerged to prescribe
how circumstances, even under static conditions, affect a couple's desired
regime of child spacing. Clearly it is still more difficult to deduce how
parents adjust their flow of births with the course of time as environmental
changes modify their reproductive goals.
As a first approximation, therefore, reproductive goals will be summarized
in terms of a desired lifetime stock of children. Accepting this working
hypothesis, economists have begun to explore parent revealed demands for
lifetime stocks of children as though conditioned by traditional determinants
of consumer and producer demand: input and output prices, income, technology,
and tastes. Ignoring radically different strategies in the timing of births,
demand for annual increments to the existing stock of children, or period
specific birth rates, should also be systematically related to revealed
demand for a lifetime stock. 3 This paper explores empirically several aspects
of the time dimension of the relationship between cohort fertility in Taiwan
and the presumed determinants of lifetime reproductive goals, namely, the
-2-
value of time of women and men proxied by their schooling, accumulated and
recent child mortality experience, and the availability of birth control
information and services. First, the accumulated stocks of births are
analyzed by age of women and regional variation in this measure of cohort
fertility is decomposed into effects operating through the age-at-marriage
and through the birth rate per year of marriage. Second, the simplest
possible stock adjustment framework is fit to the data on reproductive
stocks and flows in Taiwan to describe the dynamics of behavioral change
in a population that has experienced disequilibrating demographic, social
4 and economic change for several decades. Coefficients from stock and flow
demand equations estimated for various years are then used to appraise whether
in Taiwan these relationships are relatively constant across birth cohorts
and over time.
Several qualifications and limitations to this investigation should be
stressed at the outset, that cannot be corrected here for want of appropriate
individual panel data or analytical tools that can cope with the probable
complexity of reproductive capabilities and preferences. The most serious
limitation is the unit of analysis: large regional populations of women
born in various time periods. These aggregates are the only units for which
data are publically available on both the stocks and flows of births in
Taiwan. Investigation at the level of individuals is also imperative, per-
mitting disaggregation by women's educational attainment, a factor that
appears crucial for understanding the changing age pattern of reproductive
behavior in contemporary Taiwan. Nonetheless, despite the well known de-
ficiencies of aggregate data, it may still be fruitful to estimate behavioral
relationships at different levels of aggregation in order to document the
-3-
value and limitations of each unit of analysis; to neglect widely available
information on grounds of "principle" is hard to justify.
Aside from subjective preferences of parents for bearing their own
children, social restrictions on their exchange in most cultures encourage
5 parents to produce their own supply. Variability in supply, or the bio-
logical capacity to bear children, prevents some individuals in all popula-
tions from achieving their reproductive goals. Yet biological differences
in the supply of births do not appear to exert a dominant effect on aggregate
fertility except under extreme conditions of malnutrition and specific endemic
disease, e.g., gonorrhea. It is assumed that in Taiwan recent regional dif-
ferences in fertility are not substantially affected by such health and nu-
tritional impairments to the aggregate supply of births. 6
Most studies of the determinants of reproductive demands have dealt with
high income countries, and consequently consumer demand theory is emphasized.
In low income countries, children are more obviously a productive asset, at
least at maturity if not always at birth (Mandami, 1972; Nag, 1976).
The theory of producer derived demand for inputs might providea framework
better suited to explaining differences in fertility in developing countries.
A standard model of investment behavior in a durable input assumes that
demand is homogeneous of degree zero in all prices, holding constant the
interest rate (Griliches, 1960). But imposing this restriction appears in-
advisable in this case, for a couple's demand for children is limited both
because consumer benefits from children are probably satiable, and because
the cost of funds to invest in one's children is undoubtedly upward sloping. 7
Producer demand theory also relies on assumptions of constant returns to scale,
competitive input and output markets, and (observable) financial markets for
-4-
borrowing, none of which is appealing in the study of household demand for
children.
It is still useful to explore the stylized dynamic framework of the stock
adjustment model that has been extensively used to study demand for durable
producer inputs and durable consumer goods. The stock adjustment model
applied to reproductive behavior is not invoked here to
prescribe the path of life cycle accumulation against which reproductive
performance of a cohort can be evaluated before it reaches the end of its
potential childbearing period. This shortcoming is, of course, just another
reflection of our inability to specify determinants of the spacing of births.
Substantial differences remain to be explained across countries at one point
in time, and among countries over time, in the relative distribution of births
8 by age of mother. However, subject to identification and estimation problems
discussed in subsequent sections, information for women of a particular age
can be used to infer the current speed with which the apparent gap between current
stock and lifetime desired stock of children is being closed. If this responsP
parameter is assumed constant across a society but possibly variable over time,
such a parameter is estimable from interregional variation in age specific
reproductive behavior. Comparisons between stock and flow predictive equation
may also help us understand how the demographic transition works its way through
a population.
The paper is ordered as follows. The next section describes a few
salient features of the situation in Taiwan for which a model is soup;~t,
and relates the limitations of available data for testing an aggregate mo~pl.
The stock adjustment framework is adapted to reproductive behavior in th•, t~1 ir 0-J
section, with discussion focused on the siaplifying assumptions implied L::
this model and on the estimation problems. Rep,ional variation in cohort
-5-
stock fertility is decomposed in section four into marriage duration and
marital fertility to acquire insight into the responsiveness of fertility
and social institutions such as marriage to environmental change. The
stock adjustment model is estimated in section five and the results are
discussed further in a concluding section.
"" - .:. ~-- ,:-_ . .,. .. : .... . .,._· .: ....
-6-
II. DESCRIPTION OF TAIWAN AND AVAILABLE DATA
Demographic Transition
Mortality declined in Taiwan, notably among adults, during the period
of Japanese colonial administration of the island i.e. 1895-1945 (Barclay,
1954). Though the rise in per capita income among the Taiwanese in this inter-
war period was probably less than the substantial growth in agricultural
productivity, food consumption by the Taiwanese increased (Ho, 1966). The
more dramatic second phase of mortality reduction occurred a~er the Second
World War with the land reform and economic recovery. The greatest proportion-
ate declines were achieved among infants and children, and though the evidence
is not firm, the rural public health program, universal education and decreased
income inequality may have all contributed to this achievement. Undoubtedly,
the growth of income and personal consumption facilitated this change; since.
1952 the rate of per capita economic growth has been high by world standards,
particularly after 1962. Today the expectation of life at birth is 67 for
men and 72 for women, not far short of that recorded in high income countries. 9
Though the demographic transition began building from the start of the
century, the first indication of a decline in birth rates emerged in the late
1950 's among older women, and then only after a moderate postwar baby boom had
run its course. But in the subsequent span of twenty years, the total fertility
rate, that is, the sum of annual age specific birth rates, decreased by half
{See appendix Table A-1). This was first caused by a reduction in the frequency
of childbearing among women over the age of 30, and in the last decade the patter1
of declining birth rates gradually spread to younger women. This was accompanied
by a slow rise in the age at marriage (see Appendix Table A-2) which can be tracec
- .... _. -- ~ •..
-7-
irregularly back to the turn of the century (Goode, 1970; Barclay, 1954). As
a consequence of the separation over time of the declines in death and birth
rates, the annual rate of population growth in Taiwan increased from about one
percent in the first two decades of this century, to 2.3 percent during the
interwar period, and peaked at more than 3.5 percent during the 1950's.
Population growth has begun to ease in recent years and was somewhat less
than two percent per year in 1974 (Appendix tableA-3).
Family Planning Program
Taiwan organized and executed one of the first, most extensively studied,
and apparently effective national family planning programs in the world.
Starting in 1963 with a controlled social experiment in the city of Taichung
to determine the acceptability and effectiveness of family planning activity,
an island-wide program was expanded in several years to every township and
city precinct in the country(Freedman and Takeshita, 1969). Analyses of
regional birth rates and regional family planning activity find a strong
negative partial association between the seemingly random allocation of family
planning field workers and the level and decline in birth rates of women over
the age of 30 (Freedman and Takeshita, 1969; Hermalin, 1968, 1971; Schultz,
1969b, 1971, 1974 ). The implied effect of program personnel on birth rates,
however, diminishes from 1965 to 1968, and after about 1969 it becomes difficult
to assess whether or not the accumulated activity of the program has continued
to affect birth rates by a statistically significant amount (Schultz, 1969b,1971).
This finding can be explained in part as a natural cycle in the diffusion
of an innovation; with the introduction of distinctly superior technology
for birth control, i.e. the IUD and pill, the period of disequilibrium behavior
that follows is likely to be shortened by the subsidized dissemination of
-8-
information, services and supplies relevant to adoption. But in contrast with
the classical case of agricultural extension activities in a dynamic productive
environment, there has been only one quantum advance in birth control
technology, not a stream of improved inputs and combinations of inputs to enhance
yields and lower costs. Hence, the family planning innovation cycle is likely
to eventually meet with diminishing returns to scale (extention effort per
women of childbearing age) unless communication between generations is
absent. This tendency is already evident from cross sectional analyses of
program inputs and birth rates af'ter two years, even though the output of
services and supplies distributed to the population exhibited a more near~y
linear relationship for several years (Schultz, 1969b, 1971).
Another partial explanation for the difficulty of assessing the regional
impact of the family planning program af'ter 1968 is the limitation of the small
(361 subdivisions) units of analysis, and the uncontrolled interregional flows
of knowledge, services and users (migration). The spillover of influence of
local program activity beyond regional boundaries may have blurred the cross
sectional associations between treatments and outcomes after several years. A
similar spillover effect was thought to have been a shortcoming of the 'l'aichung
City experimental design in 1963 (Freedman and Takeshita, 1969).
Regardless, program activity in the initial years is unambigouously
linked to lower birth rates among older women, and as one might expect, the
two classes of field workers working for different government agencies appear
to be substitutes for each other in bringing about this outcome (1974a). Some
indications are found that those regions that were lagging in reducing their
birth rates toward the levels predicted by an economic-demographic model
estimated from initial period cross sectional data were regions in which the
-9-
family planning program had its greatest effect (Schultz 1?'74a).
Possibly more important than narrowing unexplained interregional differ-
ences in reproductive performance, public support for the diffusion of modern
means of birth control narrows socioeconomic class differences in contraceptivr-
knowledge and use, and thereby moderates class differentials in fert ili t.)'
that appear to especially penalize the upward mobility of the lower class
during the transitionary period of rapid population growth (Nelson, ~l. 19'71;
Freedman and Berelson, 1976). These changes in class differentials of con-
traceptive knowledge, use, and fertility are carefully documented in Taiwan
during the 1960's and 1970's (Freedman, et.al. 1974), but it remains difficult
to infer how much of these changes is due to Taiwan's Family Planning
Program.
~ducation and Fertility
I should like to interpret educational attainment as a proxy for the
nvalue of time" of men anc'I women. It is appropriate, therefore, for me to
marshall evidence of the relation between education anrl wage rates for men and
women in Taiwan. But I have as yet found no primary data on this score, and no
published analysis of education's effect on earnings in Taiwan. 10 Though this
probably reflects my inability to read the relevant Chinese literature, it does not
diminish the obvious emphasis recently given to education by the government
and the people of Taiwan. For example, from 1966 to 1974, the proportion of
men age 20-24 with some junior high school increased 32 percent, from .44
to .58, while exposure to junior high school increased 125 percent among
women of the same age, from .24 to .54. This increase in the proportion
reaching junior high school in an eight year period is all the more remarkable
when it is realized that the size of the birth cohort to educate in that
period also increased by about 80 percent.
-10-
Direct evidence, however, is available that educational attainment is
associated with reproductive behavior in Taiwan; for whatever reason. Tables
1 and 2 report 1974 birth rates, calculated by date of occurrence, for
mothers and fathers, respectively, by age and educational attainment. Three
things may be noted. A sharp reduction in total fertility rates (i.e., sum
of age specific birth rates times five) occurs among women going beyond pri-
mary school. If the distribution of education must be summarized by a single
measure, the distinctly nonlinear relationship with fertility is perhaps
better represented by the proportion continuing on to junior high school than
by an average number of years of schooling (implying a linear relationship),
or another higher cut off point such as college education, which is further
from the mode of the educational attainment distribution.
The second observation drawn from Table 1 is more tenuous, for
here the cross section of age groups is used to infer the longitudinal pace
of reproduction. In 1974, better educated women start having births at a later ag
than do less educated women, but they also appear to continue bearing children
somewhat later, into their 30's. This is a relatively new pattern in later age-
education specific birth rates in Taiwan that was less evident in 1971 or
1966 (Anderson, 1973; Taiwan Demographic Facthook, 1974, p. 15; Freedman, et.al.,
1976). To investigate these changing patterns of childbearing would require
information on stocks and flows of births by educational group, which are not
published. These changes in the timing of childbearing may explain why ear.lier
anal;yses of cross sectional changes in birth rates found that the negative
,< !
·'1
!·
Educational Level
Illiterate and those without schooling
Literate without graduating from Primary school
Primary school graduate without graduating from Junior High or Jr. Vocational School
15-19
110.5
97.2
48.7
Junior High graduate 8.0 without graduating from Senior High or Sr. Voe. School
Senior High graduate 6.6 without graduating from Jr. College or University
Jr. College & College 11.7 graduate, graduate school attended, graduate school graduate
All Educational Groups 32.2
TABLE 1
!2li-Birth Rates by Age & Educational Attainment of Mother (per thousand by date of occurrence)
-l.L-
20-24 25-29 30-34 35-39 40-44 45-49 Total Fertility rate*
315.8 223.3 82.8 33.6 9.3 1.1 3882
247.7 192.2 77. 3 28.2 10.7 1. 7 3275
244.3 235.7 95.3 34.6 9.7 1. 7 3350
136.4 221.7 96.6 29.2 5.2 1.9 2495
70.7 201.3 112.5 48.6 6.5 0.7 . 2235
37.7 163.3 101. 7 41.1 8.4 1.3 1826
183.3 219.0 91.2 32.7 9.3 1.4 2846
*Total fertility rate is five times the sum of age specific birth rates. Source: 1974 Taiwan-Fukien Demographic Factbook, tables 4 and 48 for Taiwan Area.
-12-TABLE 2
1974 Birth Rates by Age & ~9ucational Attainment of Father (per thousand by date of occurrence)
Total Educational Level l5-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 6o+ Fertility
Rate* Illiterate and those without schooling 5.3 72.4 188.7 180.6 79.9 31.5 14.2 4.2 1.5 0.3 2893 Literate 13.2 67.7 220.0 155.2 67.1 31. 8 20.5 8.9 3.9 0.8 2946 without graduating from primary school Primary school graduate 8.4 83.8 261.0 177.5 67.3 28.6 18.2 8.6 3.4 0.7 3288 without graduating from Junior High or Jr. Vocational School Junior High graduate 2.2 55.4 231.5 169.0 59.6 43.0 33.4 12.8 4.7 0.9 3063 without graduating from Junior College or University Senior High graduate 2.1 24.3 174.2 151.7 49.9 46.4 34.1 11.8 4.8 0.5 2499 without graduating from Junior College
' or University Junior College and 8.9 19.0 109.7 137.0 55.3 46.2 33.3 10.1 2.6 0.7 2114 University graduate and graduate school attended and graduate All Educational 4.6 57.0 218.4 167.0 64.4 33.8 23.5 9.0 3.2 0.5 2907 Groups
, *Total Fertility rate is defined as for women as five times the sum of the age specific birth rates.
Source: 1974 Taiwan Fukien Demographic Factbook, tables 4 and 47 for Taiwan area.
-13-
effects of women's education on birth rates were attenuated after age 34
(Schultz, 1974 .). The partial association between current births and sex
specific educational attainment may be seen more clearly when condition-
ed on the number of children already born to ~ducational groups.
The third regularity to note is the lesser, more ambiguous,variation in
birth rates with father's education (Table 2) than with mother's education
(Table 1 ). From illiterates to those with higher education, mother's total
fertility rates decline almost monotonically by 53 percent, whereas father's
total fertility rates rise 13 percent, peaking among primary graduates, and
then fall 28 percent below the level of those men with no education. Since
the correlation between husband's and wife's education is substantial in
most societies, we should expect the partial effect of women's education holding
husband's education (and earnings) constant to be even more negative, and
conversely, the partial effect of men's education to be less negative and perhaps
even positive. This result would be consistent with our expectation that the
income effect of men's earnings outweigh their price-of-time effect, but the
price-of-time effect embodied in women's value of time (education) outweigh the
income effect, reducing reproductive demands as women's education rises
(Willis, 1974; Schultz 1976 , Ben Porath 1975).
Though the advance of women in secondary schooling relative to men
has already been cited, Table 3 presents the parallel data for literacy
and higher education, and extends the data series to earlier birth cohorts.
Though women have gradually increased their literacy, as have men, the
notable advance of women into secondary and higher education has occurred
largely in the period since 1950. It may be asked, how much of the decli~e
in fertility has been simply due to the increased educational attainment of
-14-
TABLE 3
Educational Attainment at Age 20 to 24 2 by Year of Birth and Sex a
(in Percentages)
Educational Level or Above 1950-1954 1942-1946 1932-1936 1922-1926 1912-1926
Literate
Men 99 97 90 85 69
Women 96 82 66 46 26
Attending some Junior High School:
Men 58 44 27 34 20
Women 54 24 11 11 6
Attending some Higher Education
Men 16 10 s 7 6
Women 10 5 1 1 1
aThe 1950-54 birth cohort is observed in 1974, before all members may have attended a higher educational institution. The earlier cohorts are all observed as of 1966 {Census) and it is therefore assumed no differential mortality by educational level •ffects their enumerated composition at that later date. The emigration of close to a million Chinese from the Mainland in the post Second World War period augmented notably the male proportion with secondary and higher education in the cohort born between 1922 and 1926 and residing in Taiwan in 1966.
Source: 1950-1954 birth cohort--1974 Taiwan-Fukien Demographic Factbook, table 4. Earlier cohorts--1966 Taiwan Population Census, Vol. II, No. 3, table 2.
. ...._ : . ~-. , .. _ ~ .... - .: ....
-15-
women? Partitioning the change in crude birth rates into changes in women's
age composition, educational composition and a residual change within age/
education cells, it was found that 24 percent of the decline in age specific
rates from 1966 to 1974 was accounted for. by change in the distribution of
women by five educational classes (Freedman, et. al., 1976).
Child Mortality and the Demographic Transition
For reasons that may be intuitively plausible, if not derived from a
simple formal model of reproductive behavior, fertility is generally higher
in populations that experience higher child mortality rates. At the re-
gional or individual level differences in child mortality are observed to
be directly associated with differences in fertility, moderating and some-
times reversing the cross sectional pattern between mortality and surviving
family size. This evidence is strongly suggestive of a mechanism, probably
both involuntary (biological) and voluntary (behavioral) in nature, that
achieves some manner of population equilibrium given environmental health
and economic constraints (Schultz, 1967, 1976; Dumond, 1975).
But existing evidence does not explain how such modifications
in fertility are accomplished, nor how rapidly
they occur as the regime of mortality changes. Knowledge of the mechanisms
involved and of the lags in adjustment are essential to assess the duration
of the current phase of rapid population growth in low income countries, and
to appraise the gains and losses from policy interventions that seek to
improve nutrition and health, and thereby reduce mortality more rapidly.
The data from Taiwan may be useful in exploring these questions; the
Household Registry System appears to be a relatively accurate source of
current information on fertility and mortality; the 1966 Census retrospective
information from women on their number of children ever born and
.,,. ... ~ •..
-16-the survival status of their offspring is internally
consistent and plausible in all regions of the Island. It is possible, then
in Taiwan to hold constant for past child losses and examine how the recent
regional variation in child mortality is associated with current fertility.
Overview of Available Data
The unit of analysis is a highly aggregated region of Taiwan: five major
cities and 16 counties. Only for these large subdivisions are the number of
children ever born and the number of children living reported
age of woman (1966 Census). The Household Registration System has pub-
lished information since 1961 on births by age groups of mothers, and deaths
by age of the deceased. A number of assumptions are made to estimate the
stock of children ever born and the number living for earlier and later years,
using as a benchmark the birth cohorts as enumerated in the 1966 Census.11
The marital status of women is reported by age groups, and distributed
according to the year of their first marriage (1966 Census). The relationship
between mean age of a cohort and the mean age of first marriage is approximated
within each region and used to interpolate regional estimates for the standard-
ized five year age groups for which fertility data are available. Births are
not published, to nzy- knowledge, by current age and age at marriage.
Educational attainment of the population is available in various
censuses by age and sex, and is recently published from the Household Registry
system. Data on regional economic conditions are regrettably scarce for
Taiwan. A Household Income and Expenditure Survey is tabulated by regions
for the first time in 1970, but sampling variability may be a serious limita-
tion of these data as well as the lack of disaggregation by age and/or
educational attainment of household head. The unweighted means and standard
deviations of variables used is later analyses are summarized in Appendix
Table A-4.
-17-
Since the regional observations for Taiwan coincide with five cities
and sixteen less urbanized and rural areas, it could be anticipated that
relationships noted between fertility and such conditioning variables as
child mortality and schooling could simply reflect urban/rural differences.
If in fact, other environmental conditions called "modernization" or
culturally induced "norms and tastes" were responsible for urban/differ-
ences in fertility, then a causal role might be erroneously attributed
to health and education. Unless a case is made for the exogeneity of ob-
servable variables that produce the conditions of "modernization, norms,
or tastes" it is difficult to conclude that these alternative factors
are better or worse at explaining fertility than child mortality and sex-
specific schooling. It is of some interest, nonetheless, to determine
how much of the partial association between fertility and specific char-
acteristics such as child mortality and schooling are captured by the
direct admission of different levels of fertility (stocks and flows) in
urban and rural regions. To perform this test an urban dummy variable
is simply introduced into the explanatory model, even though we are unable
at this time to pinpoint precisely what objective features of the urban
and rural environment might be responsible for such shifts in behavior.
-18-
Conclusions
Taiwan was launched into the demographic transition by changes
in social and economic organization first imposed by the Japanese, followed
by heavy investments in agricultural infrastructure (Barclay, 1954; Ho,
1966). Deeper structural change in the ownership of productive assets
after the Second World War facilitated rapid industrialization and urbanization,
while policies also promoted the rapid modernization of small scale agriculture.
Costly investments in education and public health were undertaken that accelerated
the declining trend in death rates and possibly fostered labor mobility,
both developments closely associated with modern economic growth and en-
hanced labor productivity. The remarkable pace of recent economic growth
and fertility decline holds out the possibility that more could be learned
from this unusual period that would have somewhat wider applicability, and
relevance for policy; what was the role of growth in economic product,
investment in human capital, intervention to hasten the adoption of modern
birth control technology, and the peculiar social and economic institutions
of Taiwan? The available data, though exceptional with respect to aggregate demo
graphic detail, limit the goals of this study to the examination of crude
proxies for the level and composition of personal household income and relative
prices. To refine further the questions that currently occupy economic
demography it may be necessary to analyze household
economic information, which will almost certainly entail
the use of sample surveys to collect time budget data as well as income,
wealth and expenditure detail (see Kelly's paper in this volume).
-19-
III. A STOCK ADJUSTMENT MODEL OF REPRODUCTION
My objective is to estimate an adaptive model of demand for a dur-
able--children--that might clarify the process by which reproductive be-
havior responds over time to disequilibria caused by economic and demographic
change. A framework to account for both stocks and flows of births may also
provide a means for modeling the important component decisions that deter-
mine reproductive performance, namely, the timing of marriage and the
spacing of births. The standard variety of rigid stock adjustment model is
proposed as only a useful starting point for such exercises.
To simplify the task, I neglect certain aspects of the problem
that might elicit different strategies of decisionmaking in forming a family.
A couple's reproductive preferences are represented by a single-valued
indicator of their desired lifetime stock of births. Several strong assumptions
are implied. First, it is assumed that preferences among alternative family size
outcomes greater or less than the single-valued goal do not
influence reproductive outcomes. In fact, given the uncertainty that
attaches to both the biological capacity to bear children and their subsequent
survival and development, parents probably weigh the consequences of a wide range
of family size outcomes that are likely to occur with different probabilities
conditional on their behavior (Schultz, 1967). Some segments of society exceed
their reproductive goals and others fall short of theirs, possibly because
their preferences are asymmetric in the vicinity of their single most preferred
fami]ysize goal. Pioneering research on the measurement and interpretation
of family size preferences indicates that asymmetries in these preferences may
be important for understanding differences in fertility in Taiwan, at least at the
level of the individual survey respondent or across education classes (Coombs,
-20-
1974, 1976; Coombs and Sun, 1976). When regions are the unit of analysis,
within a single cultural area, this assumption may be somewhat less re-
strictive.
The second simplification is required to deal with child mortality.
The frequency of child mortality is undoubtedly affected by the availability
of household resources, production and consumption technology, and relative
prices, and it may under some circumstances even reflect allocative de-
cisions and preferences of household members, all of which are to some
degree endogenous. 11 Nonetheless, it is widely believed that regional and
time series variation in aggregate mortality rates are attributable pri-
marily to climate, public investments, available drugs and medical knowledge,
and modifications in social organization, and not due to household decision-
making. Therefore, given the scarcity of predetermined factors that are
thought to influence reproductive behavior, I shall treat child mortality 12 here as exogenous to the fertility decision.
The consequences of child mortality on reproductive goals and behavior
are' too complex to simply restate demands in terms of "surviving children!113
In the long run, as the level of child mortality decreases, the number of
births needed to achieve a given number of survivors decreases,and the
average cost of rearing a child to maturity decreases, while at the same
time all investments in the human agent, including children, appreciate
in value (Schultz, 1976). Though an economist may aspire to sort out
these offsetting supply, price, and also wealth and cross-substitution effects of
mortality on the demand for births, the essential question for population
growth is simply the overall magnitude and time path by which fertility
adapts to change in mortality (see Ben Eorath's paper in this volume).
-21-
A S~mple Framewor~ for the Joint Analysis of Stocks and Flows
With these simplifications, I assume that parents, at a particular time
t, desire a specific number of births, C~, over their lifetime. Demand for
this durable stock will depend upon what people expect of the future, and,
of course, their own preferences. The formation of expectations must be ex-
pressed in terms of current or past conditioning variables. Psychological
and economi~ aspects of habit persistence, perception, information processing,
and uncertainty are all cited as justifications for assuming the existence of
distributed lags mediating the effect of stimuli on behavior (Nerlove, 1958).
(1)
where the Zi's are M conditioning variables whose effect on C* extend for
n periods, a and 8ij (i=l, .•• , M; j=l, ••• , n) are parameters, and ut is
a residual disturbance that represents the net effect of many omitted factors
and any errors of approximation in the functional form of the relationship.
Given the central role of multiplicative interactions between births, child
survival rates, surviving offspring~ and price effects reflected in the relativ£
educational attainment of women to men, the dependent and independent variables
in equation (1) are all expressed in (natural) logarithms, unless otherwise
noted.14
Primarily for biological reasons a lapse of time is required for
the realization of desired increments to the existing stock of births,
just as technological (and economic) factors introduce lags between capital
investment decisions in plant and equipment and realized increases in
productive capacity. Though the human gestation period is only
-22-
three-fourths of one year, the median interbirth interval for couples who
report they want an additional birth immediately varies from one to three
years, depending on age of spouses, and perhaps their health and nutritional
15 status.
A conventional representation of the stock adjustment process assumes
that a proportion, o(a), of the relative difference between desired stock·
and the actual stock is delivered in each time period. For the study of
reproduction a minimum lag of a year for conception and gestation would
seem appropriate.
C - C = o(a)(C* - C ) + f(a) t t-1 t t-1 ' (2)
where 0 .< o(a) < 1, the index "a" being possibly related to a woman's age,
for reasons of biological reproductive capacity and the desired relative distribu-
tion of births over the reproductive period, and f (a) is an excess fertility func-
tion discussed below. Actually the speed of reproductive adjustment is affected b
many considerations, only the most obvious and perhaps not the most important of
which is the biological constraint imposed by reproductive potential. Given a
lifetime reproductive goal of 3 children, and a tendency to have one birth every
third year after marriage before terminating childbearing, one might expect
O(a) to be about 0.1 at the start of marriage. The relationship may deviate
from log-linearity when large increases are sought, and of course decreases
in the stock are inadmissable. These shortcomings of the stock adjustment
framework for the study of reproductive behavior are quite analogous to widely
recognized but frequently ignored defects of the framework for analysis of in-
vestment behavior. But more serious, in my judgment, is the inability to deal
explicitly with the imperfect control a couple exercises over the accumulation
of stocks.
Because of birth control failures, some women wanting no more children
have births. Consequently, even when a cohort's average number of births
equals or exceeds the average preferred number of births, some women may
I I I
I I I I I I I I
I I
I I I 1·
I I
I I
I 1.
I I
I f
l
-23-
prefer more children, and will, therefore, continue to try to have additional
births. Table 4 shows this "excess" fertility by mother's education and
age, as reported in a recent survey of Taiwan (Freedman, et. al, 1976).
The precise behavior of f(a) with respect to age is not clear, since the
proportion of women wanting no more children (col. 5 or 7) rises with age,
while their current period reproductive capacity decreases with age. To
sort out the offsetting factors that underlie f(a) requires individual
survey data on preferences and reproductive preformance, or substantially
stronger assumptions (see Lee, 1976). At the aggregate level of analysis
undertaken here, f(a) is simply interpreted as a margin of excess fertility
that cannot now be statistically distinguished from o(a)a.
The annual flow of births or the birth probability is defined,
(3)
since the stock of births are expressed in logarithms. Substituting equation
(1) into (2), a function for the growth of the stock of births is obtained.
If we collapse the expectation formation distributed lag into a
discrete lag of T years, say the mean length of the underlying distributed
lag, then either the flow of births relative to prior stock as in equation
(4) or the current period as in equation (5) below becomes a simple ex-
pression of prior stock and discretely lagged conditioning variables.
M ct = o(a)a+f(a)+ o(a) l BiZi t-T + (l-o(a))Ct-1 + o(a) ut
i=l , (5)
TABLE 4 -23a-Actual and Preferred Number of Births of Wives in 1973,
Proportion Currently Married in 1971, and Birth Rate in 1974, by Age and Education
Age and Education
Age 20-24 1. Illiterate
2. Some Primary
3. Primary Graduate
4. Junior High Graduate
5. Senior High Graduate
Total
Age 25-29 1. Illiterate
2. Some Primary 3. Primary
Graduate 4. Junior High
Graduate 5. Senior High
Graduate Total
Age 30-34
(1)
Sample Size Wives KAP-IV
1973
135
98
729
88
79
1129
248
127 792
157
149
1473
1. Illiterate 509 2. Some Primary 164 3. Primary 623
Graduate 4. Junior High 104
Graduate 5. Senior High 102
Graduate Total 1502
Age 35-39 1. Illiterate 2. Some Primary
3. Primary Graduate
4. Junior High Graduate
5. Senior High Graduate Total
497 295 502
88
52
1434
(2)
Mean Live Births
1973
2.11
1.95
1.61
1.30
.87
1.62
3.09
3.06 2.73
2.25
1.52
2.65
4.06
3.86 3.65
3.29
2.55
3. 71
4.63 4.78 4.19
3.74
2.96
4.39
(3)
Mean Pre-f erred No. of Children
1973
3.31
3.05
3.03
2.63
2.44
2.99
3.35
3.24 3.12
2.68
2.42
3.05
3.61
3.38
3.27
2.97
2.60
3.33
3.79 3. 71
3.52
3.07
2.63
3.60
(4) Differences Between Pre-ferred and
Actual
(3) - (2)
1.21
1.10
1.42
1.33
1.57
1. 37
.26
.18
.40
.43
.90
.40
-.45
-.48 -.28
-.32
.05
.38
-.84 -1.07 -.67
-.67
-.33
-.79
(5) Proportion Wives Want-ing No More
Children
1973
.341
.398
.276
.205
.165
.281
.641
.654
.562
.541
.389
.563
.841
.817
.830
.856
.794
.832
.918
.946
.902
.932
.827
.916
(6)
Proportion Currently Married
1971
.738
. 704 b
.545
.338
.171
.480
.939
.931 b
.889
.811
.691
.870
.948
.958b
.933
.910
.848
.935
.949
.953b
.925
.918
.908
.938
(7) Estimate Wanting No More a Children
(5) * (6)
.252
.280
.150
.069
.028
.135
.602
.609
.500
.439
.269
.490
. 772
.783 • 774
• 779
.673
• 770
.871
.902
.834
.856
.751
.859
2 This estimate is based on the extreme assumption that those women not currently ·married would want (more) children if they could become married.
bcategories called literate appears to refer to persons who are literate but not graduates of primary school.
(8)
Birth Probabilil
or Rate
1974
.304
.250
.249
.133
.074
.193
.217
.192
.248
.219
.209
.228
.078
.075
.098
.095
.112
.091
.034
.029
.035
.027
.030
.032
Sources: Col 1-5, R. Freedman, et. al, 1976, table 10, KAP Survey IV, 1973, Wives; Col. 6, Ibid , table 8; Col 8, Ibid table---Y and 1974 Demographic Factbook, table 18 based on date of roo-i af-T'~f-i ""
-24-
Either equation (5) or a comparable discretely lagged version of equation
(4) yields identical parameter estimates and standard errors. The high
2 collinearity between Ct ·and Ct-l yields, however, a higher R in equation
(5) and an "inflated" value of the t ratio for the coefficient on C 1 . t-
Hence, results are subsequented reported in terms of the flow equation
(4).
Commonly o(a)Si is interpreted as a short run (one year plus T) demand
elasticity of demand with respect to Zi, and Si is the analogous long run
demand elasticity. This interpretation, however, is not appropriate here
for the long run, since the value o(a) is only fixed for a birth cohort
five years in breadth. For example, if o(a) was .1, and the coefficient
estimated on Zi was .5, the short run elasticity would be .5 and the ,t-T
five year elasticity for a woman to pass through this segment of her life
cycle cannot be inferred readily from information about an age cross section.
Estimation
Even when o(a) is assumed constant, as may be tenable within a narrow
age group, the estimation of equation (5) presents problems. Many of the
omitted factors that account for the residual, ut, in the desired stock
equation (1), persist for an individual population over time or for a
cohort as it ages. The distrubances are likely to be, therefore, positively
serially correlated over time, at least toward the end of the child bearing
period, and ordinary least squares (OLS) estimates will be biased because
Ct-l will tend to be positively correlated with o(a)ut. Notably, the OLS
estimates of (1-o(a)) will tend to be biased upward (positively), and con-
verseley, estimates of o(a) are biased downward (negatively) (Nerlove,
1958; Griliches, 1960, 1961).
-25-
This simultaneous equation bias can be eliminated if the prior stock
is separately identified with additional information, or in this case,
one or more instruments are obtained that are independent of ut but are
related to C 1 • These insturments act as important identifying restric-t-
tions on this model of reproductive behavior; they determine the meaning-
16 fulness of the entire exercise.
The lagged fertility stock variable can also be replaced by its
determinants, and by repeating this substitution process until the start
of the cohort's reproductive period all lagged values of C will be elirn-
. inated from the equation. This reduced form equation would require
simplification to be empirically practical. In the case of conditioning
factors, Z's, that did not change from the start, a single long run response
coefficient could be estimated. The response to accumulated cohort child
mortality is less adequately incorporated into such a model, for in this
case the dynamic path of adjustment to the timing of the child mortality
may be important. But it would seem a useful exercise, nonetheless, to
compare the short-run response coefficients obtained from flow equation (5)
with the long-run response coefficients obtained from even a simplified
reduced form stock equation. In the next section, empirical specification
of Z and the choice of identifying restrictions are discussed that permit
one to estimate the stock and relative flow equations.
-26-
IV. DURATION OF MARRIAGE AND MARITAL FERTILITY RATE: ESTIMATION OF REDUCED FORMS
Reproductive behavior in Taiwan is first summarized by fitting reduced
form relationships for the stock of children ever born per woman by age as
reported in the 1966 Census. Within five year birth cohorts a logarithmic
specification is estimated from data for 21 administrative regions of the
. 1 d 17 is an • The following cohort specific explanatory variables are considered:
(i) the reciprocal of the accumulated child survival rate; (ii) the proportion
of women with some junior high school education as a proxy for.the value of
a mother's time; (iii) the proportion of men with the same level of schooling 18 (of the same age) as a proxy for male labor earnings; and·(iv) the man
months of family planning field worker activity in the region per woman of
childbearing age i.e., 15 to 49. All but the family planning input variable
are expressed in logarithms and derived directly from the 1966 Census.
Since cohort fertility may vary because of variation in either the timing
of marriage or the level of marital fertility, these multiplicative components
are treated as dependent variables in subsequent parallel logarithmic regres-
sions. The sum of the regression coefficients (or elasticities for those in
logs) from the component equations equals the regression coefficient from
the overall cohort fertility regression; the two way decomposition of the
logarithmic variance of fertility is thus straightforward.
If those married in a given cohort were married for the same number
of years, on average, across regions, the readily observed proportion married
at a specific age would be a reasonable proxy for the mean duration of
marriage, except for a scale factor (constant) that would change with
current age. The nearly universal exposure of Taiwanese women to
marriage, however, tnakes this assumption unsatisfactory among older women
when variation in the proportion ever married is relatively minor. For
example, after age 35, 98 percent or more of Taiwanese women have been
. d 19 marrie •
-27-
More satisfactory figures for age at marriage are obtained from 1966
Census tabulations of married women by current age and age at marriage.
How mean age at marriage is estimated from published data within regions is
explained in the second part of data Appendix B, and estimates of marital
duration are reported in Table B-4. 20 The logarithm of cohort fertility
is then linearly decomposed into two dependent variables: (1) the logarithm
of the average years of exposure to marriage per woman, and (2) the logarithm
of the residually defined annual marital fertility rate,- namely,
the number of children born divided by the years of marital exposure.
Since fertility may vary over the life cycle, the level of marital fertility
rates may be expected to reflect this and the constant terms are likely to 21 decrease among older age groups.
Cohort Fertility
Among women over age 30, when childbearing is nearly completed, the
proportion of women with some junior high school experience is negatively
associated with cohort fertility (Table 5). The absolute value of the
elasticity of fertility with respect to this measure of women's schooling
increases in the cross section to age 44, and then diminishes (later ages
not shown). The partial association between men's schooling and reproductive
performance is less uniform, though a positive partial association is
evident between the ages 20-24 and 30-39. The hypothesis that in the
postwar era the growth in men's schooling, and presumably income, is
22 associated with increased demand for children is not rejected by these data .
. After age 24 the women's education coefficient (elasticity) exceeds the men's
in absolute value and though the level of women's education is lower than
Table 5
Regressions on Cohort Fertility or Stocks: Logarithm of Children Ever Born per Women by Age in 1966a
Cohort Proportion with Some Family
-28-
R2 Age of Constant Child b Jr. High School.ing Planning Up to 1965c (SEE)d Women Term Mortality Women Men
15-19 -2.98 20.1 1.07 -1.61 -144. .3628 (6. 83) (1. 75) (1. 50) (1.17) (1.09) (.366)
20-24 -.613 22.1 -.269 .664 12.6 .6789 (4.21) (S.56) (1. 5 7) (2.33) . (. 31) ( .111}
25-29 .562 S.94 -.0886 .0542 9.21 .8062 (10. 2) (5.07) (1. 54) ( .58) (.64) ( .0403)
30-34 1.07 3.28 -.0992 .0686 -9.81 .8913 (32. 7) (5.90) (4.09) (1. 79) (1.11) (.02410)
35-39 1. 26 1. 77 -.148 .103 -17.7 .8865 (32.4) (3. 87) (6.29) (2.76) (1.69) (. 0289)
40-44 1.31 .680 -.168 .0761 -17.3 .8777 (26. 7) (1. 59) (5.01) (1.18) (1. 31) (.0366)
45-49 1. 38 -.335 -.118 -.0452 -3.33 .8422 (20.8) (.78) (3.10) (.65) (.19) (.0466)
Notes: a
b
In parentheses beneath regression coefficients are t values. Observations are 21 major subdivisions of Taiwan for which data are published in the Taiwan 1966 Census, Vol. II, 2 and 3.
Reciprocal of cohort's proportion of children everborn who are still living as reported in 1966 Census.
cMan months of family planning field worker effort expended in region through calendar year 1965 divided by the number of wom~n in the region of childbearing (i.e., 15-49) age.
dStandard error of regression estimate reported in parentheses beneath R2 .
men's, it has recently been increasing at a much faster rate than
has men's. Similar results are found for both sexes
when other levels of educational attainment are used in place of junior
23 high school.
-29-
Child mortality is positively associated with cohort fertility among
women less than age 45; after age 20-24 the magnitude of the elasticity
of fertility with respect to child mortality falls with age. The regression
coefficient on child mortality changes to a negative sign among still older cohorts
but looses statistical significance. Among women over age 39, interregional
variation in cohort fertility is insufficient to "offset" variation in
child mortality, or in other words since the child mortality elasticity is less
than +1.0, areas of relatively high fertility report relatively high
surviving fertility.Among younger aged women the reverse is noted; high
fertility areas are associated with relatively low surviving fertility, other
things equal. Problems of measurement lead one to suspect that the coefficient
on child mortality is biased in a positive direction, but the magnitude
of this bias is likely to be substantial only for the younger women. 24
The Timing of Marriage and Marital Fertility
In diverse premodern and preindustrial societies it is observed that the
age at marriage is an important regulator of lifetime reproductive performance.
To perpetuate society and maintain family lines, children are encouraged in
the face of heavy child mortality to marry and start bearing children at an
early age. This institutionalized adaptation to the regionally anticipated
level of mortality relieves individual couples of some of the burden of con-
trolling their fertility within marriage in response to actual child mortality
(Dumond, 1975). A couple's fertility might then respond to whether it ex-
perienced above or below average child losses, but this latter within-marriage
lagged response to child mortality might be difficult to distinguish with
-30-
aggregate data.
The median age at marriage in Taiwan increased from about 18 at the
turn of the century (Goode, 1970), to 20 in 1920, and to about 23 today
(Cf. Table A-2). Assuming that contemporary birth rates for married women
did not change, delaying marriage five years from age 18 to 23 implies one to two
fewer births per woman; compared with traditional cohort lifetime fertility
of five or six births, this represents a substantial reduction in cohort
fertility. An understanding of the causes for this magnitude of secular
change in the timing of marriage or even lesser differences across regions,
should be a help in explaining fertility declines.
The question I want to explore is the extent to which age at marriage
accounts for regional differences in cohort fertility, and whether these
patterns of marriage are readily explained by conditioning economic and
demographic variables that are thought to modify reproductive demands? In
the traditional Chinese family the timing of the marriage decision is, for
the most part, made by parents for their children, and relaxation of this
control is a quite recent phenomenon {Wolf, 1972). The age at marriage,
therefore, is likely to reflect the parent's perception of the benefits and
costs of earlier marriage, of which the interval to childbearing is probably
important, as well as the time parents require to accumulate a girl's dowry.
Conversely, the frequency within marriage and the lifetime number of births
may reflect to a greater degree the perceptions and interests of the younger
generation of parents. The economic incentives of a husband's and wife's
value-of-time are more likely to make themselves evident in this later de-
cision making process, though admittedly the dividing line between generations
and their respective interests is not always clear (Ben Porath, 1975).
-31-
The regressions on the estimated duration of marriage and marital
fertility rate are shown in Tables 6 and 7. The duration of marriage within
an age group is associated with the
regional incidence of child mortality among all cohorts of women over age
20. The regression coefficients from the marital duration equation for
women age 30-34 (Table 6), imply that a decline in child mortality from
15 to 5 percent, as is recorded between women age 45-49 and 30-34, is
associated with a compensating variation in age at marriage of nearly 2
years, other things unchanged. Though this estimate is probably biased upward
25 because of problems of measurement, the linkage between child mortality and
the timing of marriage deserves further study to find out why it arises, how
fast it responds to change, and whether economic and social policies can
facilitate this potentially important institutional response to diminished
1 . 26 morta ity.
The proportions of men and women with junior high schooling are not con-
sistently related to women's age at marriage, except perhaps among older
women, namely those over age 44 in 1966 (Table 6). In older groups. (not shown).
there is a slight tendency for women to marry earlier in regions where
women had more access to secondary schooling; conversely, men's schooling
is associated with somewhat later marriage among women, as is common today
in high income countries.
Marital fertility rates are not consistently associated at the regional
level with child mortality (Table 7) but a negative bias at younger ages is
anticipated. The schooling variables, that are interpreted as the value of
husband and wife time, account for much ~f the regional variation in later t
marital fertility rates; in other words, birth control within marriage is
strongly affected by schooling in the anticipated manner, with women's schoolinr.
depressing fertility and men's schooling augmenting fertility.
Table 6 -3la-
Age of Women
15-19e
20-24e
25-29
30-34
35-39
40-44
45-49
Notes: a-d
Regressions on Duration of Marriage: Logarithm of the Average Years of Exposure to Marriage per Woman by Age in 1966a
Cohort Proportion with Some Family Constant Child b Jr. High Schooling Planning c Term Mortality Women Men Up to 1965
-2.92 20.l . 775 -1.48 -137 . (6.00) (1. 57) (. 98) (. 97) ( .93)
-.177 25.0 -.356 .696 -7'.68 (. 93) (4.81) (1. 59) (1. 86). (.15)
1.37 6.11 .0676 -.199 20.2 (16.0) (3.36) (.76) (1. 37) (.90)
2.19 2.18 -.0060 -.0265 6.92 (45.0) (2.65) ( .17) (. 47) (. 53)
2.63 1.04 -.0065 -.0162 7.22 (79.1) (2.66) (.32) (. 51) (.80)
2.95 .507 .0053 -.0327 4.90 (107.) (2.11) (. 28) (.90) (.66)
3.18 .315 .0124 -.0353 4.41 (126.) (1. 94) (. 86) (1. 35) (. 68)
See Table 5.
~arital duration calculated by indirect procedure for women less than 25 years old. See Data Appendix.
R2 (SEE)d
. 3081 (.409)
.6384 (.145)
.6146 (.0626)
.5340 ( .0357)
.5287 (.0247)
.4583 (.0205)
.4627 (.0177)
fRegression coefficient for cohort child mortality is biased upward, par-ticularly for younger women as explained in ·text and footnote 25 because of measurement error.
Age of Women
15-19e
20-24e
25-29
30-34
35-39
40-44
45-49
Notes: a-d
Table 7
Regressions on Marital Fertility Rate: Logarithm of Children Ever Born per year of Marital Exposure by Age in 1966a
Cohort Proportion with Some Family Constant Child Jr. High Schooling Planning
Term Mortality Women Men Up to 1965c
-.0599 .0379 .290 -.126 -6.76 (. 58) ( .01) (1. 72) (.39) (.22)
-.436 -2.91 .0863 .0325 20.2 (7.62) (1.86) (1. 28) (.29) (1. 28)
-.805 -.169 -.156 .253 -11.0 (10.8) ( .11) (1. 99) (1.99) (.56)
-1.12 ·1.09 -.0931 .0950 -16.7 (20.5) (1.19) (2.32) (1. 50) (1.14)
-1.38 . 723 -.142 .119 -24.9 (24.5) (1.09) (4.15) (2.21) (1. 64)
-.1.64 .172 -.173 .109 -22.1 (26.5) (.32) (4.11) (1. 34) (1. 33)
-1.80 -.650 -.130 -.0099 -7. 72 (24.0) (1. 34) (3.03) ( .13) (.40)
See Table 5.
-32-
R2 (SEE)d
.4906 (.0873)
.5352 (.0435)
.2315 (.0548)
.4839 (.0399)
.6939 (.0418)
. 7785 (.0460)
.7828 (.0529)
~arital fertility rate calculated by indirect procedure for women less than 25 years old. See Data Appendix.
fRegression coefficient for cohort child mortality is biased downward, particularly for younger women as explained in text and footnote 25 because of measurement error.
-33-
Increasing both men's and women's schooling by similar proportions,
marital fertility rates decrease among women over age 34. But given the
actual proportionate changes in the last eight years (1966-1974) in men's
and women's schooling for those age 20-24 (Table 3), the regression equations
imply a 7 percent reduction in marital fertility rates for women age 25-29,
9 percent age 30-34, 14 percent age 35-39, and 19 percent age 40-44. Since
sex specific levels of schooling are not notably associated with the regional
pattern of marriage, the effect of the expansion of the educational system,
and in particular the relative gains women have made in that system in the
last 20 years, account for large decreases in cohort fertility between the
age of 35 and 49.
Two years after the start of the national family planning program there
are already indications that local program activity is beginning to modify
the regional pattern of completed fertility among older women (Table 5).
But for women 35 to 39 in 1966, only about 6 to 8 percent of their children
were born in 1965 and 1966. Thus the impact of the program on their com-
pleted fertility must inherently be marginal, and naturally this effect
operates through reducing marital birth rates (Table 7).
Tentative Conclusions
Among younger women reaching their 30s in the later 1960's, regional
variation in age at marriage appears to have over compensated for remaining
regional differences in child mortality. In regions with relatively high
fertility and high child mortality, these younger cohorts are achieving tra-
ditional reproductive goals at an earlier age than did their parents genera-
tion. If marital fertility is not excessively difficult or costly to control,
these younger women would seem likely to reduce their flow of additional births
-16-
in the decade following the 1966 Census.
Since a single cross section of a population by age provides no way
to disentangle life cycle effects from birth cohort or time series effects,
the tendency for the elasticity of cohort fertility with respect to child
mortality (Table 5) to diminish with age admits to more than one interpre-
tation.
Mortality in Taiwan appears to have declined most rapidly in two periods:
during the first decade of this century, and again from 1945 to 1955.
For women over age 44, born before 1921, childbearing was largely completed
before the second period. Moreover, many of the offspring to these older
women may have died in the dislocation and conflict of the war years, and
the aftermath of epidemics. A smaller reproductive response relative to
accumulated child losses among these older cohorts might be anticipated.
Alternatively, as a cohort advances through its life, the elasticity of
fertility with respect to child mortality may be expected to decline,because
offspring continue to die after their mother is unable to replace them with
additional births. This gradual process should be increasingly noticeable
after women reach 35 and average fecundity falls. The marked decline in
reproductive response to child mortality with increased age can, therefore,
be explained either in terms of life cycle aging or changing historical
events. It is also possible that errors in measuring child survival to a
comparable age and the possible relationship between early childbearing and
infant loss might exaggerate the positive association noted here between
fertility and cohort child mortality, especially among younger women.
In sum, fitting a simple reduced form equation for stocks of children
confirms the commonly found positive relationship with child mortality, the
negative relationship with women's schooling, and a slight indication that
-35-
men's schooling is positively related to fertility. The family planning
program inputs after two years are slightly related to lower completed
fertility among women over age 29, which replicates earlier analyses of
birth rates and family planning activity at a lower level of disaggregation
(Schultz, 1969b). The regional cohort association with child mortality is
primarily explained by the earlier age of marriage in high mortality regions.
On the other hand, the anticipated effects of regional sex specific schooling
levels on fertility is not achieved by variation in the timing of marriage,
but by changes in the rate of births per year of marriage duration. The
effect of women's schooling on marital fertility rates is negative and con-
sistent across age groups, substantially exceeding the summation of positive
men's schooling elasticities. The advance made by Taiwanese women, both
absolutely and relative to men, in gaining access to secondary schooling in
the postwar period can thus account for a substantial fraction of the con-
temporary decline in cohort fertility. If these cross sectional relation-
ships are stable over time, as will be investigated in the next section, they
also imply that the recent decline in fertility will continue.
-36-
V. A STOCK ADJUSTMENT MODEL FOR CURRENT FERTILITY IN 1967
The flow of births in 1967 as a proportion of the prior stock of births
in 1966 is the dependent variable in the simplified stock adjustment
equation (4). Using both ordinary least squares (OLS) and instrumental
variable (IV) techniques, the latter procedure being more appropriate
if C 1 is not independent of u , are shown in Table 8. In addition to t- t
the contemporaneous schooling variables for men and women, period specific
child mortality and accumulated family planning inputs are lagged two and one
year respectively. (See earlier work by Schultz (1969, 1974 ) for justi-
fication of lag structures). The lagged stock of children ever born is
identified by two instrumental variables: the cohort's prior child mortality
experience, and family planning inputs prior to 1966. The reduced form
equation that implicitly accounts for the 1966 fertility stock is reported
in Table 5.
The stock adjustment model revolves around the parameter o(a), or minus
the regression coefficient on the 1966 children ever born variable (C 1). t-
The instrumental variable estimates of this parameter in Table 8 for the
seven childbearing age groups are as follows: .07, .15, .20, .01, -.01,
-.00, and -.00. Given the low level and possibly unplanned nature of fertility
in older ages, the implied lack of discernable compensatory adjustment in
these age groups is not unanticipated. The moderate and statistically sig-
nificant level of the estimates of o(a) from age 21 to 30 does not contradict
the working hypothesis of the adjustment model over the prime childbearing
years, but these single year estimates for 1967 provide little support for
the framework at younger and older ages.
In contrast to the earlier analysis of cohort marital fertility rates,
the dynamic stock adjustment model implies compensating higher current flow
of births to women over 35 in regions where child mortality has recently
Age of Women
16-20
21-25
26-30
31-35
36-40
41-45
~6-50
Notes:
I Table 8
-37-Stock Adjustment Eguation: Relative Change in Children Ever Born in 1967a
Period Proportion with some Family Children Estimation Constant Child b Jr. High Schooling , Planning Ever Born B Methodd Term Mortality Women Men t-1 t-1 (SE
t-2 OLS .299 -1.07 -.326 • 358 34.3 -.110 .83
(2.07) ( .64) (3 .46) (2.14) (2.66) (2.91) (. 04 IV .938 -6.83 -.558 .541 52.1 -.075
(. 93) (. 73) (1. 44) (1. 38) (1.51) (. 26) (. 07 OLS .187 -.0896 -.0517 .0156 7.04 -.162 .95
(12.3) (.33) (4.44) (.81) (3. 32) (11. 5) (. 00 IV .199 -.316 -.0501 .00828 6.49 -.146
(8.78) (.76) (4.07) (. 37) (2.78) (5.59) (. 00 OLS .183 .307 -:;.0181 .00963 -.819 -.130 .66
(10.6) (1.51) (2. 38) (.82) (.56) (4.20) (.OO IV .216 .658 -.0225 .0110 .142 -.198
(. 4. 40) (1.25) (2.13) (.81) ( .07) (2.02) (.00 OLS .0568 .288 -.00955 .0170 -2.73 -.0203 • 73
(1. 54) (1. 88) (2.40) (3.52) (3.10) ( .58) (.00 IV .0286 .186 -.00791 .0167 -2.65 - .00660
(. 4 7) (.80) (1.61) (3.38) (2.93) (.11) (. 00 OLS -.00589 .257 -.00134 .00847 -1. 81 • O!HO .88
(2. 8) (3. 82) (. 53) (3. 34) (3. 39) ( .65) (.00 IV -.00937 .249 -.00101 .00829 -1. 77 .0137
(.29) (2. 78) (.30) (2.93) (3.03) (. 53) (. 00 OLS .00678 .133 -.00175 .00419 -. 779 -.00406 .78
(.74) (4.92) (1. 24) (2.38) (2. 72) (. 59) (. ()() IV -.00283 .121 -.00066 .00381 -.688 .00319
(.15) (3. 54) (. 28) (1. 96) (2.06) (. 23) (. 00 OLS .00435 .0179 -.00276 .000049 -.003 -.00315 . () 5
(3.07) (3. 71) (1. 38) ( .17) (. 062) (3.13) (. 00 IV -.00105 .0210 .000189 .000223 .022 . 000761
(.12) (2 .50) (. 24) (.46) (.25) (.12) (.on -- ···-··----- -
at or asymptotic t values are reported in parentheses beneath regression coefficients. Observations are 21 major subdivisions of Taiwan for which data are published in the Taiwan 1966 Census, Vol. II-2 and 3. The dependent variable is the difference between the logarithms of children ever born to the cohort in t and t-1, i.e., nn C - R.,n C so that the regression coefficient on the lagged stock of children "" t-1 t' is an estimate of o.
bReciprocal of child survival rate derived from period age specific death rates from birth to age 15, lagged two years, i.e., for 1965. The choice of the two year lag is discussed in Schultz 1974a.
(Notes continued)
-38-
Notes to Table 8 continued
c Man months of ef.fort by family planning field workers expended in region through 1967 divided by the number of women in the region of child bearing age (i.e., 15-49).
dRZ is inappropriate basis for comparison with IV estimates.
eOLS: IV:
Ordinary least squares estimates. Instrumental variable estimates where children ever born in 1966 is treated as endogenous and cohort child mortality and pre-1966 family planning inputs are the excluded instruments used to identify equation.
-39-
been higher. When women are completing the formation of their families,
their reproductive behavior is likely to be more sensitive to the survival
or death of earlier children. This has been found empirically in numerous
studies (Schultz, 1974 ) and in this case the short run response elasticity
is about • 2 from age 30 to 39. The magnitude of this short run response
exceeds that which could be attributed to involuntary biological feedback
mechanisms in a healthy population (Schultz, 1976).
Women's schooling is associated with lower current flows of births
among women up to age 35. For men's schooling, the positive relationship
is also apparent from 31 to 45. The men's and women's schooling elasticities
are of approximately the same magnitude among teenaged women, the women's
elasticities exceed the men's during the 20's, and the reverse is true between
the ages 30 and 44.
The intensity of family planning activity by region is associated with
a decreased flow of births among women 31 to 45, those ages where the family
planning program is widely regarded to have made its major impact (Freedman
and Takeshita, 1969; Freedman and Berelson, 1976; Schultz, 1969b, 1974a). A
very different pattern of program effectiveness emerges among women 16 to
26, where the flow of births is higher in regions that are more intensively
canvassed by the family planning program field workers. This pattern of
response among younger women in Taiwan, which I have noted before (Schultz,
1969b, 1974 ), might be explained if the preferred path to obtain the de-
sired lifetime stock of children were itself a function of birth control
technology. It was hypothesized earlier that the delay of childbearing and
the spacing of births may be a meam of reducing the likelihood of excess
fertility given the unreliability of traditional birth control measuref;.
As modern methods of fertility control become more widely accessible and
-40-·
understood, birth intervals in Taiwan may increasingly conform to the pattern
of most industrialized countries where married women frequently participate
in the nonagricultural labor force. In some of these developed countries,
the intervals between births have indeed declined in the 20th century despite
the reduction in completed cohort fertility. A corollary of this hypothesis
is that birth rates in Taiwan during the 1970's may start to decline even
among women before they reach age 30.
There are several methods for investigating the stability over time and
internal consistency of these estimated reproductive flow relationships and
the reduced form stock equations reported in the previous section. One
approach is to use the 1966 stock equation estimates in Table 5 to predict
(average regional) cohort fertility in 1971, given the observed values
of the conditioning variables observed in 1971. This exercise is reported in the
upper panel of Table 9. Apparently the 1966 relationship overpredicts declines i
fertility in the youngest ages and in the later ones. But between the ages of
25 and 39 predicted declines parallel actual reductions in cohort
fertility. The predicted declines among women reaching age 40-49 in 1971 exceed
that which might have been achieved if they had given birth to no children in
the period 1966-1971. A margin of excess fertility implied by this exercise
is not inconsistent with the evidence presented earlier in column 4 of Table 4.
The predicted reductions in cohort fertility are decomposed into the
changes attributable to changes in the four explanatory variables in the
lower panel of Table 9. The contribution of declining child mortality
appears to be of increasing importance among younger women. The reversal in
the effect of education between the ages 35-39 and 30-34 is a reflection
of the lower educational achievement in the younger age groups, who may have
been denied by the war educational opportunities compared with the older age
groups; the older cohort also includes many better educated immigrant mainland
Chinese. The postwar advance of women in the schooling system appears linked to
declines in fertility, but this effect cannot be realistically partialled out fro1
Table 9 -41-
Actual and Predi~ten C:h~npe in Samole Mean of Children Ever Born from 1966 to 1971, Based on 1Y66 Stock Esti~ates
Age of Women in 1966 and 1971
20-24 25-29 30-34 35-39 40-44 45-49 I. Relati:ve C.han~es
in Sample Mean:
Actual -.045 -.052 -.101 -.101 -.054 -.027
Predicted -.120 -.036 -.109 -.144 - .135 -.062
:r. Shares of Predicted Change:
Child Mortality -1.67 -1. 31 -.40 -.25 -.17 +.21
Women's Schooling -.86 -158 -.22 +.07 -.18 -.173
Men's Schooling
Family Planning
Total
+.qR +57 +.08 - . 18 +.01 -.20
+.55 +132 -.146 -.62 -66 -.28
-100 -100 -100 -100 -100 -100
Source: "Actual" cohort fertility based on projectipns as described in appendix B; "predicted" based on regressions coefficients in Table 5, sample means of conditioning variables in Table A-4, and projections of child mortality described in appendix B.
-42-
the educational achievements of men, which appear to be offsetting. Taken
together, education's impact varies from cohort to cohort, but promises
to increase in magnitude in the next decade among women reaching age 35.
Family planning inputs account for 28 to 66 percent of the predicted de-
clines in cohort fertility over age 30. Although these point estimates
of the impact of family planning activity are not precise (i.e., large
standard errors are associated with these coefficients) they are nonetheless
large in magnitude. As a first approximation, this exercise would suggest
that about half of the decline in fertility among women over age 30 occurrin11 i11
the period 1966 to 1971 could be attributed to reductions in child mortalitv and
changes in educational attainment, whereas the remainder is associated
with family planning activity.
Another way to test the stability of the 1966 stock equation estimates
is to reestimate these equations in 1971 as shown in Table 10. Several
changes may be noted between 1966 and 1971 coefficients. The. elasticity of
fertility with respect to child mortality has increased and become more sig-·
nificant statistically speaking after age 25; educational elasticities of both
men and women increased in later ages; the coefficients on family planning
inputs are more significant statistically and of roughly similar magnitude as
in 1966, though the level of accumulated inputs increased four fold over this
five year period. Overall, the vector of coefficients for the equation based
on 1966 and 1971 data are not dissimilar. Applying the F ratio test to the
linear restriction of coefficient equality across years one cannot reject the
hypothesis of equality in any of the six age groups at the 10 percent level. 27
Accounting for changes in stocks is predictably more difficult than )
explaining variation in levels. Reproductive behavior for a cohort approach-
ing the end of its childbearing period may be predicted with a reasonably
Table 10 -43-
Stock Reduced Form Equation for Children Ever Born 197la
Cohort Proportion with Some Family R2 Age of Constant Child b Jr. High Schooling Planning
Women Term Mortality Women Men Up to 1970c (SEE)d
20-24 -.664 21.8 .197 -.374 4.92 .5805 (3. 31) (3. 63) ( .85) (. 96) ( .17) (.123)
25-29 .539 6.99 -.0979 .0722 4.61 .6900 (5. 97) (3.33) (1.02) ( .43) (. 34) (.0539)
30-34 1.03 4.45 -.127 .107 -7.82 .7906 (15.1) (3.48) (2.12) (1.01) (.78) (.0378)
35-39 1. 28 4.40 -.127 .127 -18.5 .7854 (18.6) (3. 89) (3.58) (2.16) (1.66) ( .0397)
40-44 1.35 2.62 -.141 .108 -14.7 .8021 (18. 8) (3.03) (4. 79) (2.24) (1.25) (. 0427)
45-49 1. 35 .847 -.165 .0839 -8.39 .8534 (19. 8) (1. 39) (5.69) (1. 51) (. 72) (.0405)
50-54 1.35 -.620 -.122 -.0200 11.4 .7923 (15.3) (. 96) (3.44) (.29) (. 7 5) (.0534
See Table 5 for Notes.
,:._ v
-44-
stable equation, but year to year flows of births are more volatile and
possibly sensitive to the excessively rigid specification of
the stock adjustment hypothesis- and the functi.onal forms
adopted here. To test the stability of the stock adjustment equation, birth
cohorts are followed for eight years, 1967 to 1974 (See data appendix), and
since each year's cross section of birth rates is in some sense a new
observation conditioned by changing stocks and environmental variables, 28
the time series of cross sections are, therefore, pooled. A sample of
168 observations for each birth cohort is thereby obtained, though the
value o(a) is now undoubtedly changing as the cohort progresses eight years
through its life cycle, and its estimate should thus be interpreted with
caution. Another problem with pooling time series is the tendency for birth
rates to decline (i) for a cohort after age 25, and (ii) for most regions
within age groups over time, contributing to a pronounced secular decline
in the time series on relative changes in a cohort's stock of children.
This tendency may produce a misleading association with other strongly
trended variables, notably the past accumulated inputs of family planning
activity in a region. In Table 11 I have chosen to introduce a linear
trend in time as one method for emphasizing the cross sectional variation
about the time trend, and not the smooth time trends in variable levels. 29
These estimates are consistent, based on instrumental variables as in Table
8 to identify the influence of the lagged stock variable. 30
Coefficients in Table 11 should be compared with an analogous average
of the age specific equations in Table 8. The signs and magnitudes of the
coefficients in the two tables are not notably dissimilar, though some changes
are according to expectation. The average effect of family planning inputs
has diminished, as expected given the nature of innovational difussion cycles,
Instrumental Variable
Estimatese
Age of Women:
f,gc. 15-19 in 19'.)6 and Age 22-27 in 1974
Age 20-24 in 1966 and Age 28-32 in 1974
A~e 25-29 in 1956 and Age 33-37 in 1974
Age 30-34 in 196G and Age 38-42 in 1974
:V~e. 35-39 in 1966 and Age 4 3-11 7 i.n 197 4
Ar;e 40-44 i.n 1960 and Age l3-~2 in 1974
Notes:
Taoi.e l.l.
Sto~k Adjustment Equation: Relative Change in Stock of Children Ever Born,
Constant Period Child Tenn Mortalityb
-.853 (94)
.960 (3.19)
.782 (5.35)
.211 (5.31)
.0375 (2.10)
.0369 (1. 88)
t-2
• 710 (1.18)
-.142. (.61)
.122 (.64)
.335 (2.78)
.261 (3.91)
.192 (2.47)
1967-1974a
Proportion with Some Jr. High Schooling
Women
-.0629 (3. 27)
-.0337 (6.81)
-.00657 (1.57)
-.00794 (3.31)
-.00486 (2.38)
-.0102 (2.13)
Men
.0317 (.95)
.00752 (.85)
.00307 (.52)
.0123 (5.00)
.00701 (4.09)
.00572 (2.29)
Family Planning to t-lC
.146 (08)
.086 (.16)
-.191 (.40)
-.sos (1. 88)
-.616 (4.14)
-.361 (2.04)
Calendar Year
-19ooe
.0136 (1.10)
-.0108 (2.49)
-.00970 (3.97)
-.00224 (2.63)
-.00007 ( .18)
.00064 (1.30)
-45-
Children Ever Born t-lf
-.221 (10.0)
-.148 (6.85)
-.0510 (1.54)
-.0329 (l. 39)
-.00224 (l. 51)
-.0617 (2.06)
aAsymptotic t values reported in parentheses beneath regression coefficients. of dependent variable and e for identifying instruments and
See c for definition
bReciprocal of child survival rate derived from age specific death rates from birth to age 15 two years prior in 1965. The choice of the two year lag is discussed in Schultz 1974a.
SEE_<i
.0255
.00831
.00700
.00402
.00223
.00215
cFamily Planning inputs per woman summed to the prior year as an accumulative stock of nondepreciating knowledge.
dR2 · · · b · f . . h IV . is inappronriate asis or comparison wit estimates. (notes continued)
-46~
Notes to Table 11 continued:
eLinear time trend introduced by the variable of the last two digits to the calendar year, i.e., 66, 67 etc.
fThe dependent variable in this equation is the first difference of the logarithms of children ever born, i.e., in ct-1 - in ct' so that the regression coefficient on lagged children is an estimate of o.
gThe identifying excluded exogenous variables are the cohort's child mortality to t-2, and Family Planning inputs summed to t-2.
-47-
though the allowance of a linear trend in time may understate the program's
role in the secular down trend of birth rates. The effect of child mortality,
which is also strongly trended downward, appears somewhat larger in the
entire period 1967-1974 than it was in 1967, but a stable elasticity of
about .2 to .3 is still evident after a woman reaches age 30. The elasticity
of relative increments to the stock of births with respect to the proportion
of women with junior high school is negative at all ages, but the magnitude
of the elasticity is less than men's schooling for women between about the
ages of 30 and 40. This is again consistent with other indications that
men's schooling proxies an income effect that extends the years of child-
bearing until the woman is in her mid to late 30's, whereas women's schooling
conditional on the current stock of children exerts a dampening effect on
the flow of births throughout the life cycle. The magnitudes of the ad-
justment coefficient are better behaved except for a slight rise in level
for the oldest birth cohort. A strict interpretation of the stock adjustment
hypothesis implies reproductive goals in the range of 5 - 6 births, which
exceed survey responses,perhaps because of our inability to explicitly
identify the role of contraceptive failures. Overall, however, these
pooled results for the stock adjustment equation are more stable than we
might have anticipated given the rudimentary nature of the working hypothesis
and the limitations of the aggregate data. There is clearly an important
systematic element of feedback from past stocks to current flows of births
that should in the future be modeled with greater realism and examined in
individual survey data.
A possible shortcoming of analyses of regional data such as are used
in this study is that explanatory variables may not account for fertility
differences within urban and rural environments, but only reflect urban/rural
-48-
differences in amenities that approximately parallel without causing
differences in reproduction. As expected, a city/noncity dummy variable,
defined equal to one for the five major cities and zero for the other
sixteen regions of Taiwan, is found to be strongly negatively correlated
across our sample with reproductive stocks among older women, the simple
correlation being on the order of -.6 to -.8 in both 1966 and 1971; the
city dununy variable is less highly correlated with birth flows in 1967 among
older women, namely -.2 to -.3. Including this city dummy variable in the
birth stock equation often increases significantly the explanatory power
of the equation, particularly for older women (See appendix Tables C-1 and
C-3), whereas it is not important in partially explaining regional differences
in the flow of births (Table C-2). The only variable whose coefficient is
altered notably by this modification in model specification is that of
family planning inputs in the stock equation, where the previously noted
partial association is largely subsumed by the city/noncity distinction.
The apparent diminution of the effect of family planning on accumulated re-
productive performance is not confirmed, however, in the temporally better
specified birth flow equation for 1967, quite the contrary. Overall, it
does not appear that child mortality and schooling patterns are only a
proxy for "urbanization" and capture this alledged modernizing influence on
reproductive goals and performance. Urban/rural differences in completed
fertility remain, however, a spur to further research into reproductively
relevant characteristics of urban life or those who choose it. In 1971
women by age 40 still had ten percent fewer children if they lived in the
five major cities of Taiwan rather than elsewhere on the island (Table C-3),
after controlling for important associations with child mortality and sex
specific schooling. It is not difficult to think of many omitted variables,
-49-
particularly lagged conditions and ethnic diversity in these large popula-
tions, that might account for this residual urban/rural variation in past
reproductive performance. On the other hand, somewhat unexpectedly, one
finds that flows of births in 1967 conditioned on prior reproductive per-
formance do not differ substantially across city and noncity regional
populations in Taiwan, holding constant for the same four seemingly crude
proxies for child mortality, educational status of women and men, and prior
family planning activity.
-so-
VI. CONCLUDING NOTES AND QUALIFICATIONS
How should disequilibrium be characterized and behavior be modeled
as it adapts to unexpected changes in conditioning variables? Theories
of fertility, particularly in developing areas, are moving beyond the
widely replicated exercises of accounting for differences in lifetime stocks
of births, toward simplified explanations of how current reproductive be-
havior is conditioned by past accumulated reproductive performance and other
variables. The past matters for present behavior and future goals, as in
most areas of household behavior where the life cycle durability of decisions
is inescapable.
But past outcomes and current behavior cannot be realistically assumed
statistically independent. The identification quandary arises for lack of
time independent variables that conveniently perturb a behavioral relation
in one period but leave it untouched in the succeeding period. These exogenous
variables are hard to come by, and measure, in the household sector, and
those used in the last section of this paper, though tenable, can be easily
criticized for belonging in the current period distributed lag function.Moreover,
the stock adjustment framework is only the most simple way to deal with a
most complex process. With more and better data for cohorts over time, more
elaborate frameworks may add much to our understanding of reproductive
b l . 31 e1av1or.
Another problem arises in part because I have worked with aggregate data.
The lack of information on individual preferences, or at least the distinction
whether or not parents want more births, limits how one can treat the un-
reliability of birth control and resulting margins of "excess fertility".
Though these subjective variables add richness to a model (Lee, 1976) they
also extract their claims, for to close the system they too require ex-
planation in terms of environmentally given conditioning variables.
A third area that requires more explicit study is the age of marriage.
-51-
The timing of marriage has many implications for the allocation and accumula-
tion of resources in the household sector and the transfer of reso11rces
between f?.(_•nerations. Its effects on fertility are unmistakable. As a
starting point, decomposition of cohort fertility into an age at marriage
and a marital fertility rate may help to sort out sources of fertility change
over time, and clarify how social institutions respond to environomental
constraints. Though the required data are available in virtually every
census and survey, analysis of age at marriage remains uncommon using cross
sectional clatn. I have encountered few econometric studies of the
causes of time series chan~e in age at marriage within communities or over
generations w:ithin families, thoup,h much thinking has gone into the problem
(Goode, 1970).
The hypothesis has been advanced that education's effect on a considerable
range of household behavior can best be understood in terms of its impact on
the marginal productivity of labor, and hence the opportunity value of time
(T.W. Schultz, 1974). But education surely has other conseuqences for be-
havior and the abstraction of a single "price of time" may mislead if its
is not recognized that time is not homogeneous and perfectly substitutable
over an individual's diurnal, seasonal and life cycles. The "value of
time" hypothesis predicts that better educated women allocate more time and interes
to market oriented activities and less time to child rearing, how can one
get at directly the mechanisms by which education influences fertility ?
A variety of subtler predictions can be advanced as to how education affects
the value of time and thereby the mix of inputs used in a variety of house-
hold activities, as well as demands for final family outputs, such as
children • Are there other aspects of the husband's and wife's economic
contribution to the family that can be quantified and related to schooling?
-52-
What are the consequences of physical wealth on the household's choices?
I II • 1 II b d • Does wealth augment the strength of its owner s time-va ue ase view-
point in the allocation of family resources, or does it increase the demand
for all normal goods, particularly leisure, without introducing an off-
setting price-of-time effect? In particular, does physical wealth increase
fertility, even if it enters the family from the mother's side? What are
the limits to the family in terms of its ability to pool economic resources
and how do they change with development? Modern economic growth places a
premium on regional and occupational mobility that seems designed to erode
the economic foundation of the extended family.
Finally, the intertemporal transfer of resources is at the very heart
of the nuclear family and its relations with previous and succeeding genera-
tions. It might be postulated that the increase by half in life expectancy
in many low income countries in the last three decades would have reduced
the rate of social time preference with predictable consequences for savings
and investment behavior. The mortality reduction should also enhance the
returns on human capital relative to physical capital, shifting the balance
in family portfolios. Conversely, if income streams can be purchased more
cheaply in terms of physical assets due to rapid economic development such
as in Taiwan, will household resources allocated to enlarging the subsequent
generation diminish? The testing of many of these propositions and obtaining
a consistent set of assumptions that account for related facets of house-
hold behavior could make the "value of time" hypothesis pivotal for the
study of the household sector in low income economies.
-53-
FOOTNOTES
1The services produced by the numerical stock of children enter into
many family consumption and production activities. The output of these ac-
tivities depends on associated expenditures of time and goods, some of which
directly enhance the value of the child services such as their schooling
and health. Unfortunately, the final outputs of these family activities are
not generally observed or ascribed market prices. Therefore, the dimension
of demands for child services that is used here is simply the number of children
born or surviving to a specified mature age. See Becker (1960); Willis (1974);
DeTray (1974); Becker and Lewis (1974).
2see Zajonc, (1976) for review of evidence and a stylized interpretation
of intelligence differences by birth order and number of siblings. Lindert
(1974) presents new evidence on this pattern and provides an explanation in
terms of a mother's allocation of time among children.
3Economic models of fertility have been formulated usually in terms of
a single period static choice problem in which parents demand the optimal
lifetime stock of children for their income, relative prices and technology
(Becker, 1960; Willis, 1974). Empirical tests of this framework have examined
reproductive stocks or flows as though one was proportional to the other.
Easterlin (1968) and others have stressed a dynamic mechanism by which fertility
is adjusted in response to the gap between actual and anticipated income.
Linking cohort income deviations to relative cohort size, Lee (1975) has pro-
posed to complete the demographic feedback loop. But in the context of low
income countries or in Europe during its demographic transition, there is
surprisingly little exploration of adaptive behavioral models. Indeed the
period of demographic transition is interpreted by some as convincing evidence
-54-
that no generalizations are applicable beyond an ethnic/cultural region (Coale
1969, 1973). Given the nature of data available and the sophistication of
its analysis to date for evidence of multicausal relations, such a broad con-
clusion may be premature.
4The desirability of incorporating stocks in the interpretation of
reproductive flows was appropriately stressed by Tobin (1974). Though he is
not responsible for this application of the stock adjustment framework, his
comments stimulated my search for the stock data analyzed in this paper.
5Children are generally viewed by society as irreversible commitments
by parents and markets to exchange children are not condoned except in placing
orphans or unwanted illegitimate offspring. But in Taiwan exceptional arrange-
ments historically evolved for combining adoption and marriage to provide
parents with the opportunity to "adopt-out" girls and even boys into uxorilocal
marriages. It was connnon for a couple to adopt a baby girl who was later to
marry their son in a "sim-pua" form of marriage. This arrangement permitted
a poor couple to avoid the economic sacrifice of rearing a girl to marriagable
age, and assured the adopting couple a loyal and servile daughter-in-law
(Wolf, 1972, chapter 11). Over half of the marriages in the Taipei area prior
to 1925 did not require the transfer of a young women into her spouses house-
hold (Wolf, p. 171). Adoptions are today 50% more common for girls than for
boys, and equal about three percent of the births in 1970-72. There are other re-
flections of the lesser demand for girls (p. 60)including the reported historical
practice of infanticide among girls (p. 54). Contemporary analysis of birth
rates reveals a greater reproductive replacement response when a male infant
dies than when a female infant dies (Schultz, 1969), and male offspring pre-
ferences are well documented in contemporary Taiwan surveys (Coombs and Sun,
1976).
6s ee Tabbarah (1971) and Easterlin (1975).
-))-
In fact, difference in health
conditions may affect the rate with which cohorts achieve their lifetime
reproductive goals, but this effect is thought to be secondary to demand
factors. To the extent that health-related supply limitations were posi-
tively associated with child mortality rates, the estimated partial association
between child mortality and fertility would be biased in a negative direction
by the omission of supply limitations or their determinants. There are still
instances in which far less healthy and more malnourished populations, such
as exist in Sub-Saharan Africa and Bangladesh, might display regional differences
in fertility which are attributable to differences in reproductive capacity
or supply (Chen and Mosley, 1976). See also Easterlin, Pollack and Wachter's
paper in this volume.
7 Children as a producer durable expand the household's budget constraint
(See Kelly's paper in this conference), but not indefinitely, given imperfectly
elastic supplies of complementary inputs. For example a small farmer might
be able to borrow to buy additional land for his sons to farm, but the capital
market may not view sons as the least risky form of collateral and thus require
an increasing risk premium on such a loan.
8see U.N. (1965) for evidence of dissimilar age patterns of births
across countries and over time. As yet no characterization of optimal child
spacing strategies has gained wide acceptance, though models are implicit
in several studies (Sanderson and Willis, 1971; Heckman and Willis, 1975 ).
9Barclay (1954, pp. 154-165) suggests life expectancy increasinv, from
25-29 around the turn of the century to 40-45 by 1936-1940, but surprisirgly
little reduction occurred in infant mortality. Infant mortality declined
from levels of 160 around 1940 to 32 by 1960 and 14 in 1971-t. (197~~iwan_
Fukien Demographic Factbook). Thou~1 some understatement of mortality exists
and some transfer of infant deaths to the second year of life prohably persi~ts
-56-
in the registry, these errors are urilikely to alter the noted trends or under-
mine confidence in interregional variation in registered vital rates. (1964
Taiwan Demographic Factbook).
lOI have since found a reference by Chien-shen Shih (1976, p. 296) to
a mimeographed study on "Rates of Return to Education in Taiwan, Republic of
China, July 1972" by K.G. Gannicott for the Ministry of Education. Gannicott's
estimates of the social rate of return to education, as cited by Shih,
are 27, 12, 13, 13, and 18 percent per annum for primary, junior high, senior
high, senior vocational and university education, respectively. It is unclear
whether returns are calculated separately for men and women, or what secular
growth in real labor incomes is assumed which adjusts upward the cross
sectional age differences to obtain a synthetic estimate of longitudinal
(cohort) returns to education.
llTh . di d i 1h A d. f ese assumptions are scusse n ' e Data ppen ix; except or a
variety of smoothing procedures to interpolate individual years values, the
primary assumption is that internal migration is not selective with respect
to women according to their fertility, and that age specific mortality does
not differ by the child's mother's age.
12In 1950, for example, multiplying out age specific survival rates for
Taiwan, the life table probability for a live birth to reach age 15 was .84
and having reached 15 the chance of reaching age 30 was .95. In 1960 these
survival rates had increased to 193 and .97 respectively, and by 1974 they
stood at .96 and .98. See Taiwan Demographic Factbooks for 1964, Table 11,
and 1974, Table 34.
h l
-57-
13 An obvious approach for dealing with child mortality is to choose
a threshold age at which to measure "surviving children"; an age before
which most child mortality occurs and beyond which survival prospects are
favorable (see previous footnote). Though arbitrary, this procedure appears
at first to be an improvement over assuming that parents formulate their re-
productive goals in terms of live births, valuing all the same regardless
of survival status (See O'Hara, 1972 on problems of evaluation and summation).
But if a multiplicative model of demand for births is assumed, an:d child
survival rates are an explanatory argument, terms can be rearranged and the
hypothesis tested directly whether demand for "surviving children" is indeed
perfectly inelastic with respect to child mortality · However,
imposingthe "surviving child" hypothesis on the demand equation not only
loses information, it makes it more difficult to interpret parameter estimates
across birth cohorts of women for whom child survival is inherently observed
to different threshold ages.
The equation estimated later in Table 5 can be written
ln CEB : a+ Sln(CEB/CA) + yln SF + ~ln SM+ £FP + u,
where CEB is the number of children ever born per women, CA is the number
of those children still living per women, SF and SN are the proportions of women
and men with some junior high schooling, FP is family planning inputs per
women>u is a normally distributed constant variance disturbance, and a,S,y,6
and £ are estimated response elasticities. Rewritten in terms of the number
of living children per women, one has
B 1 y o £ ln CA = a S + <-Thu (CEB) + S ln SF + S ln SM + S FP + u/S.
But if estimated in this form the high definitional collinearity of CA and
CEB makes interpretation difficult as would the admission of errors in
measuring cohort fertility. The "surviving child" hypothesis could be rigidly
tested in this context by determining if the coefficient 6n CEB were actually
zero, i.e., S = 1.
-58-
14 The measures of education and the inverse of the child survival rate
(either based on cohort experience or recent period specific rates) are
essentially proportions and are not unreasonably specified in the double
logarithmic or constant elasticity form. Family planning effort, on the
other hand, is an absolute measure of inputs up to the previous calendar
year per potential recipient, and in this case the exponential functional
form has the appeal of permitting the elasticity to rise or, more likely,
fall with the scale of inputs. The predictive power of the cohort fertility
equation was also increased slightly when family planning inputs were
specified in a form that required the fertility response to inputs to
approach an asymptotic limit, such as l/(l+FP).
15stochastic models of the biological components to this interbirth in-
terval are compatible with the distributed lag framework, modified to allow
the lag structure to be unimodal and skewed to the right, such as the log
normal in excess of the minimum gestation period. See Sheps and Menken,
1974, and Potter, 1975.
16Another estimation approach was explored for three age gr~ups (30-44).
An interative maximum likelihood procedure would choose a value of 6, obtain
OLS estimates of 0 where the dependent variables is C -- 6C and iterate on .... t ' t-1'
6 to maximize the predicted fit for Ct. Similar parameter estimates were
obtained, but t ratios were generally increased, particularly for women's
schooling and child mortality.
-59-
17Barclay (1954, pp. 248-254) in his classic study of Taiwan up to
1945 disparages that there is little variation in the prewar high fertility
level between rural and urban areas. He notes further that "it has not
been possible to find any evidence of association between fertility and
other recorded types of behavior of rural Taiwanese by Districts" (townships)
whereas "the strongest spatial pattern of fertility was the sectional one,
viewed in prefectual units" (counties examined here). This puzzles him
since Taiwan was recently settled and did not appear to have evolved
distinct regional cultural traditions. Earlier, however, Barclay did note
that the prefectual spatial arrangement of mortality and fertility is "some-
what the same." Though Barclay does not indicate what recorded types of
behavior were investigated, it seems unlikely that he looked at child mortality,
women's schooling, or women's labor force activity outside of agriculture.
18 The husbands of a cohort of women are, on average, older. But without
information from the Census on exactly how much older, the educational attain-
ment of men of the same age has been used here. The estimated relationship
should not be confounded greatly by this relatively uniform error in measure-
ment. The tendency would be for the estimated coefficient to be biased toward
zero from that for actual husbands who are older and generally less well edu-
cated than the age group used here.
19cohort fertility can also be decomposed into the proportion married and
the number of children born per married woman. Since it is unconnnon to find
census tabulations by age at rnsrriage and current age and common to encounter
tabulations by marital status and age, it is worth noting that this less
satisfactory decomposition reveals roughly the same results as does the
duration of marriage decomposition (Table 6 and 7). Particularly among the
younger women for whom the proportion married appears a reasonable proxy
for marital duration the associations with child mortality are notable.
-60-
Similarly, the education variables are more important in the regressions on
children born per married women.
20 Divorce is rare but not absent from Taiwan, but remarriage appears to
occur promptly (Barclay, 1954, Chp. VIII). Presumably an increasing share
of the time since first marriage is spent without a husband as the cohort
ages due to widowhood. This effect may be impounded in the fertility re-
gression constant term among older women. Regional and time series varia-
tion in marriage proportions and the frequency of divorce has been attributed
to sex ratio differences in Taiwan (Goode, 1970, pp. 289-316), but this
endogenous aspect of the problem is not treated here.
21 Standardizing these marital fertility rates according to a given
reproductive schedule is not necessary. There is obviously a parallel
between this form of cohort fertility decomposition and the procedures
adopted in the Princeton European Fertility Study (Coale, 1969, 1973). Since
their methodology relies on regional period specific births and indirectly
standardizes these for age and marital status composition of the population,
a Hutterite fertility schedule is used to weight the age distribution. In
the case of Taiwan there is little evidence in cohort data of the tradeof f
found by Demeny (1968) in marriage and marital fertility indexes, but further
research on this issue is needed.
22 Though our expectation is that income from the husband's earnings
increases the demand for children, other things equal, this result does not
follow from the simple d,emand theoretical framework, without additional
assumptions (Schultz, 1974b). Therefore, a: two tailed t-test of significance
would seem appropriate in evaluating the male schooling regression coefficiPnts.
23The proportions literate, primary schooled (or more) and senior high
schooled (or more) were used in place of th · · t e proportion wit l so,me _iunior hi.vb
school (or more) as measures of schooling/wage rates. They were slightly less
successful in explaining fertility, but similar age patterns of regression
coefficients were obtained.
-01-
24since it is likely that higher fertility
is associated with earlier childbearing, the children of mother's (or a given
age) may be somewhat older, on average, in regions where fertility is higher.
Being older, child survival would be lower, even if age specific death rates
are uniform across regions. This inability to measure child mortality to
the same age level imparts an upward bias to the estimated elasticity of
cohort fertility with respect to cohort child mortality. In addition, some
evidence suggests that child mortality is greater among the offspring of
very young mothers. (See also fn. 25) • There are· clearly many difficult to
disentange life cycle relationships among age at marriage, fertility and
child mortality, and this investigation deals only with a few preliminary
indications of such relations.
25As indicated in footnote 24 the relationship between marriage d11ration
and child mortality is undoubtedly overstated (positively) since women who man- i .~,.::
earlier had their children earlier, on average. Their children were therefor·· 1 .J ck r
at the time of the Census and had experienced more mortality risks. This
effect, however, diminishes markedly as the cohort of women age and their
youngest child outgrows the period of heaviest mortality. But earlier born
children were also exposed to the earlier and undoubtedly higher infant and
childhood death rates. This time series effect and the duration effect
would both tend to bias upward the partial association between marriage
duration and child mortality. Techniques proposed by Brass/Sullivan/Trussell
to estimate life table mortality rates from retrospective survey inform~t ion
on child survival rates should help to mitigate the bias arising from the
greater age of offspring of mothers that married at a younger age.
-62-
26coale (1973, p. 57) is convinced that the timing of marriage, at
least in Europe, does not respond to reproductive goals, and thereby directs
his attention to changes in. marital fertility as a precursor to the demographic
transition. "Few couples marry at 25 instead of 24 because of a calculation
that they will have one birth less; whereas the practice of contraception or
abortion is directly aimed at fewer births". Both components of cohort
fertility may bear further examination by behavioral scientists to better
understand the demographic transition, even in the European setting.
27The F statistics with 5 and 37 degress of freedom are .98, .45, .52,
1.55, 1.59 and .95 for the age groups 20-24, 25-29, 30-34, 35-39, 40-44,
and 45-49, respectively. At the 10 percent confidence level, one could
reject the hypothesis of coefficient equality if the F(5,37) exceeded 2.01.
28child mortality is observed in each year in each region through the
age of 15. The measure of schooling of men and women, i.e. the proportion
with some junior high school, is interpolated between the 1966 Census and
tabulations published from the household registry system in the Factbooks of
1973/1974. These changes would primarily reflect cohort migration among
regions. Family planning inputs are accumulated per potential recipient.
A slightly better fit to the data is obtained if the family planning inputs
are transformed to l/(l+FP), which implies more sharply diminishing returns
to scale approaching an asymptotic limit as past inputs accumulate. The easier
to interpret linear
and 11, however.
exponential specification is retained in Table 8
29 Another procedure is to include a vector of dummy variables for
all but one calendar year to account for yearly changes in flow~ without
restricting the trend to be linear. Th. is more flexible procedure reduces
degrees of freedom, but implies similar results.
-63-
30As in Table 8 the two variables used to identify the lagged endogenous
stock variable are the cohort's child mortality experience, lagged two years,
and family planning inputs, lagged two years. Note that the cohort's child
mortality is derived initially from different data than the period specific
child mortality level that enters directly into the stock adjustment equation.
The former is based primarily on retrospective child survival as reported in
the 1966 Census and adjusted over time by period specific death rates, as
discussed in the ~ata appendix. The latter period specific variable is
obtained from the region's age specific death rates (for all cohorts) two
years prior to the dependent variable birth rate.
31 One step would be to estimate continuous distributed lag structure~>
rather than the discrete two year lag in mortality and the one year lag in
family planning inputs. Another step would be to model the innovation
(birth control technology) adoption mechanism along the lines proposed by
Welch (1970), allowing for more flexible substitution possi.Lilities between
classes of family planning workers (Schultz, 1969b) and interaction effects
between the woman's (and men's ?) schooling and the application of fam:lly
planning extension activity. When I explored such interaction effects
in the context of the stock adjustment equation (Table 8 and 11), the interaction
coefficients were positive (as expected if woman's schooling substitutes
for extension effort in diffusing modern birth control technology), but
not quite statistically significant from age 35-44. The coefficients
(and their t ratios) for the cross product terms were for age 35-39, 29.l
(1.00), age 40-44, 20.0 (1.33).
-64-
Anderson, J.E., "The Effects of Marital Fertility, Nuptiali ty and Educational Attainment on Fertility Change in Taiwan, 1966-1971", Taiwan Population Studies working paper No. 25, University of Michigan, 1Q73 (see also Studies in Family Planning, _i: 3, March 1975).
Barclay, G.W., Colonial Devel..2Pment and Population __ ~ 'l'aiwan, Princeton, H.J.: Princeton University Press, 1()5lt, reissued, Port Washinp;ton, N. Y.: Kennikat Press, 1972.
Becker, G.S., "An Economic Analysis of Fertility", in Demo~aphic anc'l. Ec_..2!1omic_ Change in Developed Countries, Universities-National Bureau ~onference 11, Princeton, N.J.: Princeton University Press, 1960.
Becker, G. S. and H. G. Lewis, "Interaction between Quantity and Q.uali ty of Children", in (ed.) T. W. Schultz, Economics of the Famiq, Chicar;o: University of Chicago Press, 1974.
Ben Porath, Y. and F. Welch, Chane~ Child Trai_~s~LChoice of ..Yamily Size, R-1117, Santa Monica, CA. The Rand Corp., Dec. 197?.
, "Fertility and Economic Growth: Some Micro-Economic Asp"rts" Discussion Paper 756, Falk Institute for Economic Research, Jerusalem Israel, October 1975.
Coale, A. J., "The Decline of Fertility in Europe from the French Revolution to World War II", in (eds.) S.J. Behrman, L. Corsa, and H. Freedman, Fertility and Family Planning, Ann Arbor, Mich.: University of Michigan Press, lGr/1 •
_____ , "The Demographic Transition Reconsidered'', paper presented at tr,e International Population Conference of the IUSSP, Leige, Belgium, 1973.
Coombs, L. C., "The Measurement of Family Size Preferences and Subsequent ~,ertility", Demography_, 11: 11, Nov. 19(h, 587-611.
, Are Cross Cultural Preference Comparisons Possible? A measurement-------theoretic approach, IUSSP Paper Ho. 5, (n.d./1976), Leige, Belgium.
------, and T.H. Sun, "Family Composition Preferences in a Developing Culture: 'rhe Case of Taiwan, 1G73", Taiwan Population Studies Workinr; Paper No. 30, April 1976, Taichung/Ann Arbor.
DeTray, D.N., "Child Quality and the Demand for Children", in (ed.), 'r.w. Schultz, Economics of the Family, Chicago: University of Chicago Press, 1974.
Demeny, Paul, '1Early Fertility decline in Austria-Hungary: A Lesson in Demop:ra1,~: i · Transition", Daedelus, 97, Spring 1968, 502-522.
Dumond, D.E., "The Limitation of Human Population: A Natural History", Scie~_c_e_, ~87, 28 February, 1975, 713-721.
Pasterlin, R.A., "An Economic Framework for Fertilit:r l\naJysi;;:, Family P~_?.nn~~. 6_:1, ~-'.arch, 1"75, 5li-(l.
-65-
--------, Population, Labor Force_ an fl _I~C2_n_g_ _~3-~nf"~ i~ _Eco_n_o_mic Growth, New York: National Bureau of Economic Hesearch, 19613.
Freedman, Ronald, "Norms for Family Size in Und.erdevel oped L:ountries", Proceedings of the Royal Statistical .?ociety, B 15'), i0r;3, 2;,n-245.
--------' and ,J. Y. Takeshita, Fami~Y...Xlan_r:!_i_n_~ _ig_~ra3=_w_~_: ___ An Experiment_ in Social Change, Princeton, N.J.: Princeton University P'.~ess, l'l6</.
~---~--~ , L.C. Coombs and M.C. Chang, "Trends in Family Size PrF!ferences and Practice of Family Planning: Taiwan, 1()65-1970'', Studies in Family Planning, }:12, Dec. 1<!7~, 281-296.
---------' L.C. Coombs, M.C. Chang, and T.H. Sun, ''Trends Family Size Preferences, and Practice of Family Planning: Studies in Family Planning, 2_:9, Sept. 1974, 270-288.
in Fert i .I ity, 'I'aiwan l<J(,')-i-1'1·<,
and B. Berelson, "The Record of Family Planning Programs", Studies in Family Plannin&, .z.:1, January 1976.
-----, M.B. Weinberger, T.H. Fan and S.P. Wei, "Recent Trends in Fertil:itv and in the Effects of Education on Fertility in Taiwan, 1961-1974", Taiwan Population Studies Working Paper No. 31, Ann Arbor, Mich., April 1976.
Goode, W.J., World Revolution and Family Patterns, New York: The Free Press, 1970.
Griliches, Zvi, "The Demand for a Durable Input: Farm Tractors in t11e United States, 1921-1957" in A.C. Harberger (ed) The Demand for Durable Goods, Chicago: University of Chicago Press, 1960.
----------~, "A Note on Serial Correlation Bias in Estimates of Distributed Lags", Economet!ica, ~:l, January 1961.
k J J d R J Willis "Estimation of a Stochastic Model of Reproduc-tinn Hee man, . . , an · • • > d · An Econometric Approach," in (ed. ) N. E. Ter lee kyi, Household I ro uc t J(~i:!_
and Consumption, Studies in Income and Wealth, No. 40, New York: National Bureau of Economic Research, 1975.
1 i Of the Effects of Acceptance on Fe rt il it':''. Hermalin, A.I., "Taiwan: An Area Ana ys s Studies in Family Planning, No. 33, August 1968, 7-11 •
, "Appraising the Effects of a F~mily Planning Program ~-~~A-r_e_a_l~Analysis", Taiwan Population Studies Working Paper, No.
versity of Michigan, 1971.
through an 14, Uni-
"Empirical Research in Taiwan on Factors Underlying Differences in Fertllity", in Economic Aspects of Population Growth, (ed.) A.J. Coale, International Economics Assn., Halsted Press, New York, 1975.
-66-
Ho, Yhi-Min, Agricultural Development of Taiwan, 1903-196Q_, Kingsport Tenn.: Vanderbilt University Press, 1966.
lnstitute of Economics, Academia Sinica, Conference on Population and Economic Development in Taiwan, Dec. 29, 1975 - Jan. 2, 1976, Taipei, Taiwan, 1976.
Lane;, .Jlga, Chinese Family and :~os_i~!_l, New Haven: Yale University Pwess, l))J.
Lee, R.D., "Fertility, Age Structure arid Income in the United States, 1947-1974" paper presented at Econometrics Society Third World Confress, Toronto, August, 1975.
, "Marital Fertility in the U.S.: 1949-1q74:, mimeo Universit:r of ----Michigan, February 197fi.
Levy, Marion, The Family Revolution in Modern China, Cambridge Mass: Harvard University Press, 1949.
Lindert ,p. II. , "Fertility and the Microeconomics of Inequality", Discussion Paper, University of Wisconsin, Madison, Wisc., 1974.
Liu, P.K.C., The Uses of Household Registration Records in Measuring the Fertility Level in Taiwan, Economic Papers No. 2, Institute of Economics, Academica Sinica, Taipei, Taiwan, 1967.
Mamdani, Mahmood, The Myth of Population Control, New York: Monthly Review Press, 1972.
Mueller, Eva,"Economic Motives for Family Limitation: A Study Conducted in Taiwan" Population Studies, ~:3, Nov. 1972, 383-403.
Mosley, W.H. and L. Chen, "Health and Human Renrodl1ct1"on i·n D 1 · ·· r eve oping Countries , Johns Hankins University, April lrj7" ('.".'limeo).
Nag, Mani, "The Economic View of Children in Agricultural Societies: A Review and a Proposal," in Culture, Natality, and Family Planning , (eds.) J.F. Marshall and S. Polgar, Monograph 21, Chapel Hill: University of North Carolina, Carolina Population Center, 1976, pp. 3-23.
Nelson, R.R., T.P. Schultz, and R.L. Slighton, Structural Change in a Developin~ Country, Princeton, N.J.: Princeton University Press, 1971.
Nerlove, Marc, Distributed Lags and Demand Analysis of Agricultural and Other Commodities, USDA Agricultural Handbook No. 141, Washington D.C., 1958.
_______ , "A Note on Error-Components Models", Econometrica, March 1971, 79: 1, 383-396.
-------' "On Lags and Economic Behavior", Econometrics, March 1972, 40:1, pp. 221-251.
and T.P. Schultz, Love and Life Between the Censuses: A model of family decision making in Puerto Rico, 1950-1960, RM-6322, Santa Monica, CA: The Rand Corp. Sept. 1970.
-67-
O'Hara, D.J., Changes in Mortality Levels and Family Decisions Regarding Chil~re~, R-914, Santa Monica CA: The Rand Corporation, 1972.
Potter, R.G., "Changes of Natural Fertility and Contraceptive Equivalents", Social Forces, 2i:l, September 1975, 36-51.
Hutstein, S.O., "The Influence of Child MOrtality on Fertility in Taiwan", Studies in Family Plannin~, .2_:6, June 1074, 182-180.
Sanderson, Warren and R.J. Willis, "Economic Monels of Fertility: i)ome Examples and Implications", New York: National Bureau of Economic Research, ( mimeo) , 1971.
Schultz, T. p. , A Family Planning Hypothesis: Gome .l!;mpirical .l!;Vidence tram Puerto Rico, RM-5405. Santa Monica, CA: The Rand Corporation, 1967 ·
, "An Economic Model of Family Planning and Fertility", ~f Polit. ~~~~E-c_o_n-.-, 77:2, March/April 19hoa, 153-180.
The Effectiveness of Fa.mU:Y___!'laE_n_:h_I!lt in_.J:..aiw1!:!l, P-hof;o, Sant a Monica, CA: The Rand Corporation, 1969b.
, Evaluation of Population Policies, R-643, Santa Monica, CA: ----~--.,~ The Rand Corp., June 1971.
, "Birth Rate Changes over Space and Time: A Stud::' of :ai wan", in ___ E_c_o_n_o-mics of the Family, (ed.) T. W. Schultz, Chicago: Uni vers1 ty of
Chicago Press, 1074 (also R-1070, '!'he Rand Corp. September 1°7~).
"Interrelationships between Mortality and Fertility", (ed.) R. G. Ridker, Population and Deve_loumen~ _Y._!i_e_~earc_l!__!_or Selecti ~e Interventions, Baltimore, MD: Johns Hopkins University Press, 1 76 •
( d ) Economics of. the Family, Uni versi t.v of Chicago Press, Schultz, T. W. e . , -· ··· Chicago 1974.
Sheps, M.C. and J.A. Menken, Mathematical Moriels of C?nception and Bii:_th, Chicago: University of Chicago Press, 1973.
Shih, Chien•sheng, "The Contribution of Education to. Economic Development in Taiwan," in Conference on Population and Economic Development in Taiwan, Taipei, Taiwan: Institute of Economics Academica Sinica, 1976, pp. 287-314.
Sun, T.H., "The Prospects for Population Growth: Socio-economic Implications for Taiwan", Conference on Population and Economic Development in Taiwan, Taipei, Taiwan: Institute of Economics, Academia Sinica, 1976.
Tabbarah, R.B., "Toward a Theory of Demographic Development, Econ. Devel.and Cultural Change, 19:2, January 1971, 257-277. ..
Taiwan Demographic Factbook, various years, 1961-1974, Department of Civ.i] Affairs and Ministry of Interior, Taiwan Provincial Government, Taiwan.
-68-
Taiwan Provincial Institute of Family Planning, £..~~_R_l_~nE_ing R_~_~r:_ence_
Book, Vol. I-XIII, Taichung, Taiwan, 1969-1975.
Taiwan Dept. of Budget, Accounting and Statistics, Report of the. Surve_L_ of Family Income and Expenditure, various years, 1966-1974, Ta iwa11 Provincial Government, Taiwan.
Taiwan Report of the Taiwan and Fukien Area ~_r_ulation and Hot~jn~~ Cens1~~ of 1966, Vol. II, Taiwan Province, No. 2, Marriage Status _c_i_f_ the_ Population and Women's Fertility.
· raiwan Report of the Taiwan and Fukine Area Population and Housing Cens~~ of 1966 9 Vol. II, Taiwan Province, No. 3, ·I'he Level of Education of the Population.
'.l'obin, James, "Comment" in (ed.) T.W. Schultz, Economi_c_s of the Family, Chicago: University of Chicago Press, 1974.
United Nations, Dept. of Economic and Social Affairs, The Determinants and Con sequences of Population Trends, Population Study No. 17, New York, 1953
United Nations, Dept. of Economic and Social Affairs, Population Bulletin, No. 1963, Fertility, New York, 1965.
Williams, A.D., "Fertility and Reproductive Loss", Unpublished Ph.D. dissertation, University of Chicago, Dec. 1975.
Willis, R.J., "Economic Theory of Fertility Behavior", in (ed.) T.W. Schultz, Economics of the Family, Chicago: University of Chicago Press, 1974.
Wolf, A.P., "Adopt a Daughter-in-Law, Marry a Sister: A Chinese Solution to the Problem of the Incest Taboo", American Anthropologist, 70:5, October 1968.
Wolf, Margery, Women and the Family in Rural Taiwan.! Stanford, Calif.: Stanfor University Press, l972.
Zajonc, R.B., "Family Configuration and Intelligence," Science, 16 Apri1 197n, 192: 4236, pp. 227-236.
Appendices
Appendix A-- Additional Statistical Tables
Appendix B-- Procedures for Regional Cohort Projections
Appendix C-- Regression Tables with City-Noncity Dummy
-68a-
Births Registered in Yeara
1949
1950 1951 1952 1953 1954
1955 1956 1957 1958 1959
1960 1961 1962 1963 1964
1965 1966 1967 1968 196'.1
1970 1971 1972 1973 1974
TABLE A-1 Age Specific Annual Birth Rates for Talwan: 1949-1974 -69-
(Births per Thousand Women of Child-bearing Age)
Total r~umb.:r of Births Per 1000 Women of Soecific Ages Fertility ----------
Rateb 15-19 20-24 25-29 30-34 35-39 !10-44
5900 61 241 290 263 186 111
6030 61 246 297 269 191 112 7040 68 287 349 311 ·226 132 6615 53 272 342 294 220 113 6470 48 265 336 292 218 108 6425 48 263 334 292 218 104
6530 50 273 341 295 219 103 6505 51 2M 340 296 222 105 6000 45 249 325 275 197 92 6055 43 248 336 281 199 90 5990 46 258 334 270 190 86
5750 48 253 333 255 169 79 5585 45 ?.48 3112 246 156 71 5465 45 255 338 235 145 65 5350 41 252 337 231 139 60 5100 37 254 335 214 120 52
4825 36 261 326 195 100 41 4815 40 274 326 188 91 38 t,220 39 250 295 158 70 28 4325 41 256 309 161 68 26 4120 40 245 298 151 63 23
4000 40 238 293 147 59 20 3705 36 224 277 13!1 51 16 3365 35 208 257 117 41 13 3210 33 203 250 105 37 12 301+5 34 197 235 96 35 10
aEirths are attributed to the year they arc registered.
bThc total fertility rate is five times the sum of the a?;e specific birth rates. Perhaps because of rc·_mding the totals do not aluays add up.
c
l.S-49
'i-7
30 35 29 27 26
25 23 17 14 14
13 10 10 10
8
6 6 4 4 4
3 3 2 2 2
c Births by age of mother divided by the mid-year esti.rn.--1tc of the number of wc1Den of that age.
Source: 19119-196!1, _!2.cmographJc· E<!_(:_!_T'.:_rwk 1%l1 196ft-1974, _1=!emor.;raph_i:..<:_X<~~J.:__Boo_!~.J-974
-70-
TABLE A-2
Percentage of Women Ever Married in Taiwan by Age, 1940-1974
Period and I P E R C E N T 0 F AGE GR 0 UP Source 15-19 20-24 25-29 30-34 35-39 4o-44 45-49
1940 Census I 29.5 84.4 95.9 98.3 98.8 99.2 99.4
1956 Census ,. 11. 5 70.6 q5.2 97.9 98. 5 98.7 99.0
1966 Census 8.6 5(). 5 <)2.9 98.1 98.9 99.1 99.1
1970 Census I 7.2 49.7 91. 3 97.8 98.8 98.8 98.8 f.
1974 Household 5.8 44.1 84.9 95.4 97.0 97.6 97.4 Registry ...
Sources: 19lio Census Table 13, p. 54. 1956 Census Table 17, p. 265. 1966 Census Vol. II, No. 3, Table 2, p. 125 1970 Census, extract, Table 9, p. 135. 1974 Taiwan-Fukien Demographic Factbook, Table 9, p. 366.
-71-TABLE A-3
-·--------------------------------------Period Natural Rate of Crude Vital Rates Life ExEectan1 Birthl-Dea:u;- I Population Infant Death Male Femal1
Increase (percent per year) (per Thousand Population) (per thousand (years at bir ·
---------···----------live_ bir_ths)
1906-1920 l.02a 42 31 172 25-30
1921-19110 2. 32 a 115 22 155 35-115
1<147-llf4G 2.5 4o 15 b b b
1()50-1954 3.58 45.9 10.1 37 56.1 Go.2
1955-1959 3, l19 42.9 8.o 37 60.5 (i5,9
1960-19611 3.08 37.2 G.4 30 62.7 68.ri
1 CJ65-lf:lhQ 2. 1,4 29.7 5,3 22 64.4 70.0
iq70-1974 2.01 24.8 li. 8 16 tS6.5 71.8
a,l'aiwanese geometric rate of growth between censuses (Barclay, 1054, p. 13).
b . not available.
Sources: 1906-1940, Barclay, 1954, pp. 13, 161, 241. 19117-1949, 1959-1961 Household Registry St_?.-tistic::;~f Taiwan, Table 1. 1950-196h, 1972 Demographic Factboo~'I'aiwa:q_, Table 1, 1970 Demographic
Factbook, Table 15. 19IS5-19711, 1974.'hlwan-Funkien Demogr~~ic !_a_ctbook, Tables 61, 711 and '(h.
-72-
TABLE A-4
Means and Standard Deviations of Variables Used in Regressions on Fertility*
Variable
1. Children ever born 1966: c66
2. Log (c66 )
3. Log (C6/C66)
4. Log (Reciprocal of cohort child survival 1966)
5. Log of child death rate 1965 +
6. Log (proportion women with Jr. High School)
7. Log (proportion men ~ith Sr. High School)
8. Family planning per woman to 1965
9. Family planning per woman in 1966
15-19
.0622 (. 0285)
-2.86 (.410)
.781 (.0920)
.0238 ( .00817)
20-24
.853 (.157)
-.174 ( .175)
.296 (.0297)
.0262 (.00718)
A G E G R 0 U P 0 F W 0 H E N
25-29
2.50 (.204)
.912 ( .0818)
.105 ( .00743)
.0352 (.00931)
30-34
3.95 (. 253)
1. 37 (.0653)
.0362 (.00522)
.0514 (.0120)
35-39
4.93 (.366)
1.59 (.0766)
.0197 (.00440)
.0744 (.0164)
40-44
5.44 (.483)
1.69 (.0934)
.00431 ( .00181)
.110 (.0221)
45-49
5.65 (.558)
1. 73 (.105)
.00056 (.00028)
.152 (.0277)
.0521 (. 0112)
not age specific variable, i.e., identical for all age groups
-1.16 (.339)
-.692 ( .184)
.00161 (. 00063)
.00268 (. 00081)
-1.55 (.355)
-.870 (.217)
-2.19 (. 442)
-1.25 (.259)
-2.42 (. 4 77)
-1. 39 (.300)
-2.36 (. 511)
-1.14 (.314)
-2.50 (.586)
-1.15 (.301)
not age specific variable, i.e., identical for all ages
not age specific variable, i.e., identical for all ages
* The standard deviations of the variables are reported in parentheses beneath the means• Values are unweighted over sample of 21 regions.
-2.89 (. 713)
-1.43 (. 386)
+ The natural logarithm of the reciprocal of the product of the age specific rates within the region from birth to age 15.
DATA APPENDIX B: PROCEDURES FOR REGIONAL COHORT PROJECTIONS
Consistent Estimates of Birth Stocks and Flows
First, the proportion married by single year of age j, in the ith
region is defined
. 1m .. (66) = (p .• (66) - s .. (66))/p .. (66) ]+ Jl. Jl. Jl. Jl. i=l, ••• ,21
j=lS, ••• ,49
-73-
where p refers to all women and s those single or never married according
to the 1966 Census (Vol. 3, Table 2).
In years after 1966, information is annually published from The House-
hold Registry on proportion of the population ever married by five year age
groups. Using the individual year population totals for eaeh subsequent
year, t, a predicted proportion married is calculated; for example, for
age 15 to 19:
19 19
19Ml5,i"(t) = [/~ p .. (t) f; m .. (66)]/ l p .• (t). j=lS Jl. Jl. j=lS Jl.
I then define alpha (a) as a marriage deflator for each region, year and for
the seven five-year age intervals for which household registry data are
available.
where 19m15 ,i(t) is the actual proportion of women between the ages of
15 and 19 registered as ever married in year t. If the age weights
had not relatively changed, and age specific marital rates declined over
time, the a's would presumably increase and exceed unity after 1966.
The estimated proportion of women married by single year of age, in
calendar year t, can then be expressed as follows for, say, age 18
as follows:
-74-
The second step is to estimate birth rates for women by individual
ages, though birth rates are reported from The Household Registry only
by five year intervals. Using initial arbitrary estimates of marital
fertility reported in Table B-~denoted as F., that are not untenable at J
the national level for the base year of 1966, a similar procedure of
deflation is performed to obtain estimates of individual year birth rates
that are roughly consistent with the changing age composition of regional
populations, marriage patterns and age-aggregated birth rates as recorded
in the h9usehold registration system by date of registration. The base
year estimate of the birth rate would become:
19 19 19Bl5,i(t) = [ l p .• (t) *·+1tn .. (t)*F.]/ l p .. (t)
j=l5 Ji J J,i J j=lS Ji
and a birth rate deflator for the seven age intervals is defined as Beta (6),
where 19b15 Ct) is the registered birth rate for women of age 15 to 19
in region i in calendar year t. Similarly, a final estimate of the birth
rate for women of individual ages is obtained:
Women are then followed by individual years in the 1966 Census age
distribution, attributing to them their estimates birth rate in subsequent
years, in addition to the number of children already born as reported in
the 1966 Census. These single year of birth cohorts are summed into
five year age-groups each calendar year and observations are constructed
to follow through time the aging regional cohort, neglecting for the
effects of internal migration. Thus the 35 to 39 year old women in
Table B-1
Initial Estimate of Marital Fertility Rate by Age of Woman Used to Calculate Beta
Initial Relative Age of Woman Value of Fertility
14 or less 0
15 .100 16 .200 17 .250 18 .300 19 .350
20 .400 21 .425 22 • 450 23 .425 24 .400
25 .375 26 .350 27 • 325 28 • 300 29 .275
30 .250 31 .225 32 .200 33 .175 34 .150
35 .125 36 .100 37 .090 38 • 080 39 .070
40 .060 41 .050 42 .040 43 .025 44 • 010
45 .007 46 .005 47 .003 48 .002 49 • 001
50 or more 0
-75-
region 2 (Ilan Hsien) had 5.11 children on average in 1966 and in 1971
were estimated to have at age 40 to 44 5.34 children ever born. In
contrast, those women age 35 to 39 in 1971 had on average only 4.63
· children ever born.
Survival of Cohort's Living Children
-76-
The mortality rates for each region and year are read or calculated
for infants and children age l to 4, 5 to 9, and 10 to 14. Based on
nationa.lr levels of age specific mortality between 1965 and 1970, the
mortality rates for children age 15 to 19, 20 to 24, and 25 and over
are arbitrarily assumed to be proportional to the death rate for children
10 to 14 in the region, where the factor of proportionality is 1.73,
2.55 and 2.90, respectively. Mortality among older offspring is
relatively low and intrinsically of less interest to us here because an
increasing proportion of their mothers are no longer of childbearing age.
To survive the cohort's living offspring to the next period one needs to
know the age of their children. The total number of living children for the
seven standard (five year) age intervals of women is obtained by region
fztom the 1966 Census (Vol. 3). Initially, I assume the proportion of
those.living children in each current age group is as arbitrarily reported in
Table B-2, chosen to be roughly consistent with national birth rates in
the early 1960's. But if absolute differences in infant mortality remain
substantial by region, as is the case in Taiwan in the 1960's, it seems
appropriate to use regional household registry birth rates in 1966 to
estimate directly the number of infants (age 0) by region and age of
mother, and use the relative proportions in Table B-2 to distribute only
living children age one and over. For example, women 20 to 24 in region 1
Age Group of
Mothers
15-19
20-24
25-29
30-34
35-39
40-44
45-49
50-54
,:._ v
Table B-2
Initial Proportions of Children Living by Current Age, and Age of Mother in 1966
Age of Children 0 1-4 5-9 10-14 15-19 20-24
.545 .455 0 0 0 0
• 313 .624 .063 0 0 0
.149 .525 .305 .021 0 0
.066 • 321 .414 .186 .013 0
.033 .148 • 319 .341 .150 .009
.014 .064 .166 .295 . 316 .137
.004 .023 .074 .161 . 289 • 307
0 0 .025 .073 .162 .291
,:-. v
-77-
25 or more
0
0
0
0
0
.000
.142
.449
,:_ v
-78-
(Taipei Hsien) registered a birth rate of .287 in 1966, and reported in the
Census .938 children living. Ignoring infant mortality, the balance
of children living, .65~ are distributed between the age groups 1-4 and
5-9 in proportion to their cell values in TableB-2,or 90.8 and 9.2 percent,
. 1 1 respective y. Clearly, further refinements and more extensive checks for
consistency could be introduced by using information on registered birth
and death rates in prior years (Maurer and Schultz, 1972, p. 8), but these
errors need n~~ jeopardize our objective of obtaining estimates of cohort
accumulative fertility and surviving offspring}
Overall, the procedures described above should not introduce rela-
tively large errors, unless the Census and registry systems are incompatible.
The least satisfactory assumption is probably that embodied in the uniform
age specific marital fertility schedule. There has been much change in
age specific marital birth rates in recent years in Taiwan. As the age at
marriage has increased, the marital birth rate increased among those who
still got married early, between the ages of 15 and 19 (Anderson, 1973).
This may have been partially due to changes in the age composition of
married women in this interval (becoming older) but would also be consistent
with a selectivity process by which those getting married are increasingly
1The convention is followed of subjecting the calendar year's regis-tered births to mortality of that year on December 31 before obtaining the population of living children for the next year. Hence, the Census at the end of November 1966 is taken as a year end population total, in which the births are assumed to occur on the last day of the accounting period. The 1967 year end total of one-year olds will then be the 1966 registered births diminished by the 1967 infant mortality rates. The small error in measurement introduced by this convention in the initial year, therefore, will not be accumulative, though it will overrepresent infants in these demographic accounts.
2 Methods could be applied at the regional level based only on current period fertility and mortality schedules by age, see p. 8 of K. Maurer and T.P. Schultz, A Population Projection Model, R-953, Santa Monica CA: The Rand Corp., August 1972. Using more information about recent levels of birth and death rates would seem superior, though perhaps more complicated.
fecund, either because they were pregnant before marriage (perhaps
unusually fecund for their age) or inclined to start immediately their
childbearing, and hence marrying at an atypical young age. Whatever
-79-
the cause, these changes in marital fertility rates are not allowed to
modify the relative shape of the schedule but only displace the schedule
up and down uniformly over the five year age intervals. This in fact
may be a poor approximation for how marital fertility schedules have been
changing in different regions of Taiwan. However, since our primary
interest attaches to the behavior of older women, this defect in my
calculations may not, hopefully, be a serious shortcoming for this
analysis.
Estimates of Duration of Marriage
The 1966 Census is tabulated by region, for women by five current
age intervals, and for those ever married by several age intervals of
first marriage (see TableB-3). Neglecting the intervals between dissolu-
tion of marriages and remarriage, on which there is no information, these
Census tabulations can be used to approximate the average number of years
elapsed since first marriage for each current five-year age group of
women.
There are three qistinct issues: (1) estimating the mean current
age in the age intervals reported, (2) estimating the mean age at first
marriage in the age at maPriage intervals, and (3) interpolating age at
marriage for the standard five year age intervals to complement other
published data. In the first case, the mean age within current age
intervals can be directly calculated for ever-married women at the
regional level from single year age distributions by marital status
Age at First Marriage
12-14
15-19
20-24
25-29
30-34
35-44
35-54
35 and more
Table B-3
Form of Age at Marriage Tabulations in Taiwan 1966 Census
Current Ase of Married Women 12-24 25-34 35-44 45-54
x x x x
x x x x
x x x x
x x x
x x x
x
x
Source: 1966 Taiwan Census, Vol. 3, Table 6.
-80-
55 and more
x
x
x
x
x
x
-81-
(Vol. 3 , Table 2 ) • In the second case, two different procedures are
used. Among women age 25 or more, the majority of the cohort is married.
An estimate of the average age at marriage (AM) for those married at each
current age is calculated as follows:
j=l, ••• ,5 k = 1, ••• ,n
where the regional subscript is suppressed, j refers to current age
interval, and k to age at marriage interval. A relation over age cohorts
between AM and current age would be affected by the tendency of older age groups
to have had more years to get married, increasing with age the average
reported age at marriage, other things being equal. Also, there is some
evidence that the median age at marriage has decreased among more recent
birth cohorts during the 20th century (Goode, 1960; Barclay, 1954). Linear
regressions are fitwithin regions to the four current age groups over age 24 with t
mean age at marriage expressed either in arithmetic of logarithmic form
as a function of the current mean age and an intercept. The arithmetic
form accounted satisfactorily for the secular upward trend in age at
marriage and was used to interpolate values for five year intervals
over age 25, i.e., ~oefficients of determination were between .8 and .9.
For younger women, a different approach was needed, given the form
of the age at marriage tabulations (see Table B-3).The working assumption
is that cross sectional differences in the proportion married at different
single ages represents the marrying fraction of a stable "synthetic" cohort
that begins to marry at age 12. For women currently aged 15 to 19, for
example, the average duration of marriage is then approximated as
19 1 r r 1=15 i=l2
(m. - m. )*p *(1-i)) I J J-1 1
19 l P1 '
1=15
-82-
where the regional subscript is suppressed, and m. and p. refer to single J J
year married proportions and female populations of exactly age j. For
women currently 20-24, the summations over the 1 index runs from age
20 to 24.
To determine if the first method of direct observation yields similar
results to the second, comparisons are possible only over the entire
current age interval from 15 to 24 and later age groups. The two resulting
estimates of marriage duration for age 15 to 24 are correlated at .99 over
the 21 regions. At later ages the two approaches are, as one might antici-
pate, less highly correlated.
The first direct interpolation method is used to obtain the estimates
of the average duration of marriage or years of exposure since first
marriage for the age cohorts older than 24 in 1966. The second synthetic
cross sectional method is used to obtain the duration estimates for the
two younger birth cohorts in 1966. The natural logarithm of this marital
duration variable is reported in Table B-4 by region along with the
logarithm of the cohort's birth rate per year of exposure to marriage.
The sum of these logarithmic variables is, of course, the logarithm of
the cohort's average number of birth per woman.
TABLE B-4
Logarithms of Estimated Marital Duration in Years and Marital Fertility Rate Per ¥ear -83-
A g e o f W o m e n lJ-1~ 25-2'.I 30-34 40-44 45-49
Region Duration Rate
Current LU-24
Duration - Rate Duration Rate Duration-- Rate ~ Duration Rate Duration Rate Duration Rate
1. Taipei Hsien -2.10382 -0.37907 0.55780
2. Ilan Hsien -2.32702 -0.31383 0.45386
3. Taoynan Hsien -2.19788 -0.30315 0.52947
4. Hsinchu Hsien -2.53190 -0.44011 0.27~96
5. Miaoli Hsien -2.95563 -0.37978 0.15718
6. Taichung Hsien -2.79355 -0.32774 0.22025
7. Changhwa Hsien -3.20471 -0.38523 0.09867
8. Nantou Hsien -2.57412 -0.39012 0.42442
9. Yunlin Usien -3.05406 -0.34214 0.42273
10. Chiayi Hsien -2.79355 -0.41044 0.36615
11. Tainan Hsien -2.85851 -0.39600 0.33541
12. Kaohsiung llsien -2.40954 -0.33907 0.48844
13. Pingtung Hsien -2.36250 -0.33907 0.54631
14. Taitung Hsien -1.46114 -0.43598 0.93131
15. llualien Hsien -1. 80229 -0. 40681 0. 76965
16. Penghu Hsien -2.22124 -0.40681 0.51707
17. Taipei City -2.87916 -0.14497 0.10938
18. Keelung City -2.31328 -0.29855 0.58543
19. Taichung City -2.97814 -0.07310 0.21069
20. Tainan City -2.58319 -0.06786 0.15804
21. Kaohsiung City -2.j3319 -0.19582 0.46793
-0.60008 1.72252
-0.56378 1.82714
-0.59850 1.75780
-0.53876 1.52539
-0.51614 1.69647
-0.55726 1.72383
-0.55769 1.76350
-0.61280 1:75533
-0. 6 3939 1. 76606 -0.59859 1.71474
0.56633 1.71127
-0.60049 1.69939
-0.64095 1.73837
-0.69679 1.87130
-0.62858 1.89169
-0.67512 1.68380
-0.50213 1.58571
-0.54755 1.74900
-0.47258 1.63904
-0.53521 1.59571
-0.63304 1.61767
-0.81382 2.36912
0.88363 2.42640
-0.81288 2.38573
-0.62567 2.25035
-0.81278 2.35351
-0.83792 2.37108
-0.89020 2.39278
-0.82081 2.38276
-0.81746 2.39439
-0.78803 2.36939
-0.77974 2.36981
-0.77058 2.35495
-0.78945 2.36853
-0.77446 2.43966
-0.84556 2.45088
-0.72545 2.34374
-0.86141 2.30452
-0.80420 2.38227
-0.80583 2.33181
-0.82723 2.31230
-0.78990 2.31519
-1.01742 2.75858 -1.20307 3.03816
-1.01091 2.79850' -1.16805 3.06907
-0.99612 2.76848 -1.15355 3.04460
-0.86078 2.66619 -1:06082 2.95903
-0.95674 2.74529 -1.10062 3.02657
-0.98711 2.76076 -1.14050 3.04046
-1.01548 2.77602 -1.16066 3.05240
-1.00161 2.76533 -1.15637 3.04136
-1.00955 4.77728 -1.16172 3.05348
-0.98332 2.76169 -1.14882 3.04272
-0.99729 2.76347 -1.15425 3.04519
-0.97507 2.74756 -1.13297 3.02875
-0.98359 2.75209 -1.13727 3.02864 -0.94375 2.79982 -1.09924 3.06405 -0.98566 2.80739 -1.14138 3.06966 -0.89526 2.73788 -1.02009 3.01985 -1.09198 2.71839 -1.32313 3.01025 -1.01284 2.76695 -1.23370 3.04407 -1.02575 2.73715 -1.22409 3.02477 -1.04113 2.72546 -1.23271 3.01696
-1.05075 2. 72211 -1.25518 3.0105~
-1.43261 3.25641
-1.32586 3.28181
-1.34002 3.26075
-1.25094 3.18525
cl.25766 3.24586
-1.30656 3.25878
-1.31516 3.26870 -1.31001 3.25744
-1.30157 3.26967
-1.31312 3.26185
-1.31108 3.26475
-1.30340 3.24798
-1.30868 3.24504
-1.28396 3.27285
-1.30299 3.27724
-1.21238 3.23956
-1.56729 3.23589
-1.42552 3.26083
-1.42552 3.24787
-1.42930 3.24239
-1.45753 3.23410
-1. 64611
-1.51797
-1.52457
-1.44524
-1.43801
-1.47185
-1.48560 -1.48560
-1.43490 -1. 45404
-1. 44204
-1.49287
~l. 46 783
-1. 49504
-1.50264
-1.40605
-1. 77588
-1. 73593
-1. 59262
-1. 58972
-1. 64437
-84-
Age of Women
15-19
20-24
25-29
30-34
35-39
40-44
45-49
Notes:
a-d
Table C-1
Regressions on Cohort Fertility or Stocks: Logarithm of a Children Ever Born per Women by Age in 1966
(Including Urban-Rural Component)
Cohort Proportion with Some Famiiy Plan-Constant Child Jr. High Schooling ning up to · Urban-Rural
Term Mortality Women Men 1965c Variablee
-2.53 21.5 1.41 -1.57 -125. -.343 (3.99) (1.86) (1. 79) (1.14) (.94) (.99)
-.551 22.1 -.232 .662 14.2 -.035 (2.30) (5.40) (1.11) (2. 25) (.34) (. 33)
.722 5.83 -.079 .155 16.0 -.0826 (8.23) (5.53) (1.52) (1.62) (1. 21) (2. 20)
1.19 3.32 -.0681 .0952 -1.65 -.0592 (25.0) (7 .18) (2. 97) (2 .87) (.21) (2.86)
1.37 1. 76 -.0973 .0959 -7.33 -.0691 (29.8) (4.89) (4.01) (3.25) (. 83) (3.28)
1.42 .704 -.124 .0805 -6.59 -.0759 (24. 3) (1. 94) (3.82) (1. 48) (.56) (2. 72)
1.45 -.237 -.0998 -.0193 4.45 -.0599 (17.5) (.56) (2. 57) ( .28) (.25) (1.43)
See Table 5
R2 (SEE)d
.4022 (.3665)
.6812 (.1138)
• 8534 (. 0362)
.9297 (.0200)
.9339 (. 0027)
.9180 (. 0309).
.8612 (.0452)
e Urban-rural variable equals 1 for 5 cities (Tainan, Taipei, Keelung, Taichung and Kaohsiung), 0 for 17 rural areas.
,:-_ ~
.,
Table C-2 -85-Stock Adjustment Equation: Relative Change in· Children Ever Born in 1967a
(Including urban-rural component)
Period Chi£-d Proportion with Some Family Children Urban-R2 Age of Estimation Constant Mortality Jr. High Schooling Planningc Ever Born Rural
Women Methodd Term t-2 Women Men t-1 t-1 Variable (SEE)d
16-20 OLS .308 -1.07 -.315 .354 34. 7 -.111 -.00902 .8306 (1. 99) (.62) (2.83) (2.05) (2. 5 7) (2. 82) ( .21) ( .0453)
IV .818 -5.87 -.536 .515 4.85 .0456 .0137 (1.33) (.99) (1. 81) (1.66) (1. 94) (.25) (.20) (.0661)
21-25 OLS .180 -.0818 -.0558 .0157 6.81 -.162 .00421 .9568 (9.23) (.30) (4.06) (. 80) (3.09) (11. 3) ( .60) (. 00737)
IV .190 -.282 -.0545 .00923 6.31 -.148 .00434 (7 .57) (. 70) (3.81) (.41) (2.64) (5. 83) (.60) (. 00763)
26-30 OLS .185 .312 -.0182 .0107 -. 731 -.132 -.000884- .6618 (7. 53) (1. 46) (2. 30) (.76) (.45) (3. 87) ( .15) ( .00517)
IV .240 .749 -.0238 .0191 .921 -.222 -.00620 (4.21) (1.61) (2.19) (1.01) (. 37) (2.46) (. 72) (.00633)
31-35 OLS .0416 .253 -.00993 .0159 -2.96 -.00962 .00223 .7452 (;92) (1.52) (2.42) (3.01) (3.02) (.24) (.60) (. 00315)
IV .0594 .311 -.0106 .0164 -2.93 -.0254 .00157 (.84) (1.28) (2.31) (2.98) (2.95) (. 41) (. 37) (.00317)
36-40 OLS -.0344 .215 -.00139 .00746 -2.02 .0293 .00343 .9059 (1. 38) (3.25) (.59) (3.10) (3.99) (1.59) (1. 87) (. 00161)
IV -.0244 .236 -.00202 .00792 -2.07 .0219 .00303 ( .60) (2. 50) (. 65) (2. 79) (3.92) (. 73) (1. 36) (.00162)
41-45 OLS .00589 .133 -.00174 .00415 -.789 -.00354 .000138 .7843 (.50) (4. 71) (1.19) (2.24) (2.57) (. 43) ( .13) (. 00101)
IV -.00125 .126 -.00115 .00385 -. 769 .00148 .000474 (. 05) (3.39) (. 46) (1. 79) (2.40) (. 08) (. 30) (.00102)
46-50 OLS .00420 .0181 -.000274 .000036 -.00700 -.00308 .000035 .6581 (2. 4 7) (3.54) (1. 32) ( .12) (.12) (2. 73) (.18) (.00195)
IV -.00516 .0247 .000274 .000126 -.0170 .00329 .000463 (. 27) (1.54) (. 28) (.22) (.15) (. 25) ( .50) (. 000352)
-86-Table C-3
Stock Reduced Form Eguation for Children Ever Born 1971 a
Age of Women
20-24
25-29
30-34
35-39
40-44
45-49
50-54
Notes: a-d
Constant Term
-.582 (2.49)
.556 (5.11)
1.10 (14.3) .
1.36 (17. 7)
1.47 (23.1)
1.47 (23.0)
1.46 (14.5)
See Table 5.
e See Table C-1.
... ···-·· ,:·. v
(Including Urban-Rura~ Component)
Cohort Proportion with Some Family Plan- Urban-. child Jr. High Schooling ning up to Rural b Women Men 1970C Variablee Mortality
20.9 . ·;211 -.347 11.1 -.0707 (3.34) ,, (l.05) (.87) (.36) (.71)
6.83 --~0850 .0753 6.28 0.0139 (3.06) 0: (. 79.) ( .43) (.41) (.29)
3.82 --~.106 .153 -.617 -.0584 (3.06) ; (1.:85) (1.50) (.06) (1.79)
3.74 ·~· -.0865 .145 -8.09 -.0680 (3.40) i. (2.22) (2.63) (.70) (1.93)
2.06 -.0725 .117 -.686 -.104 (3.10) (2.50) (3.24) (.07) (3.66)
.534 -.104 .0860 3.92 -.0978 (1.10) (3.53) (1. 98) (.40) (3.33)
-.640 -.0871 .00916 17.8 -.0875 (1.07) (2. 33) (.14) (1.23) (1. 92)
.... ···-·· ,:._ v ... ···-·· ,:._ v
R2 (SEE)d
.5942 (.1244)
.6917 .(.0555)
.827/i ( .0355)
.8280 (.0367)
.8955 ( .03203)
.9158 (. 0317)
.8331 (.04942)
Top Related